Drug Discovery AI

Superluminal Medicines chases the same target class as Nabla Bio and Antiverse and attacks it from the opposite direction. Those two design antibodies against G protein coupled receptors. Superluminal designs small molecules against them, which is the older and harder computational problem on a class where roughly 70 percent of the more than 800 known receptors remain undrugged despite around 35 percent of all marketed drugs acting on the family. The company launched from Boston in August 2023 with a $33 million seed led by RA Capital Management, joined by Insight Partners, Nvidia and Gaingels, and closed a $120 million Series A in September 2024 with the same lead, adding Catalio Capital Management, Eli Lilly and Company and the law firm Cooley as investors. Total disclosed funding is $153 million. Cony D'Cruz is chief executive. It styles itself the Membrane Company. The platform is called Hyperloop and the components are named rather than gestured at: structure based drug discovery, protein dynamics modelling, generative chemistry and machine learning, in silico pharmacokinetic and toxicology prediction, and experimental validation, arranged as a predict, design and test architecture. The stated aim is to model the specific structural change that produces a therapeutic effect and design selectively for it. Unusually for this lane, the company also names the third party software it runs on, using Schrodinger LiveDesign for molecular design and data analysis and CDD Vault for assay data management. Schrodinger is itself indexed here. The commercial event that defines this record came in August 2025, when Eli Lilly signed a collaboration worth up to $1.3 billion covering undisclosed receptor targets in cardiometabolic disease and obesity. The structure is described in more detail than most: upfront and near term payments, an equity investment, development and commercial milestones, and tiered royalties on net sales, with Lilly taking exclusive rights to develop and commercialise compounds arising from the work. Six small molecule programmes were reported as of the Series A, and a wholly owned candidate outside the Lilly collaboration was expected to enter human trials during 2026. The Lilly relationship is unusually entangled and a reader should hold all three strands together. Lilly invested in the Series A, then became the company's largest commercial partner twelve months later, and Superluminal's registered address is the fourth floor of Lilly Gateway Labs on Necco Street in Boston. Its investor is also its partner and its landlord. None of that makes the validation false, and a sophisticated buyer paying twice is meaningful evidence, but it is not arm's length and should not be read as though it were. One practical warning for anyone verifying the pipeline. At least one commercial drug pipeline database attributes an unrelated compound and its clinical results to this company. Nothing Superluminal has disclosed supports that attribution and it should not be carried forward.

Lila Sciences makes the largest claims of any vendor in this index and publishes the least evidence for them. Both halves of that sentence are load bearing, and a reader should hold them together. The company was founded inside Flagship Pioneering's own laboratories in 2023, spent two years in stealth, and emerged in March 2025 with $200 million in committed seed capital. Flagship is the venture creation firm behind Moderna, Generate Biomedicines and Tessera. Geoffrey von Maltzahn, a Flagship general partner, is chief executive; Noubar Afeyan, Molly Gibson, Jacob Feala, Alexandra Sneider, Ben Kompa and Scott Robertson are named among the founding team; the geneticist George Church is chief scientist and Andrew Beam is chief technology officer. Sites are in Cambridge Massachusetts, San Francisco and London. A Series A of $235 million led by Braidwell and Collective Global followed in September 2025, with reported totals near $550 million at a valuation above $1.3 billion and Nvidia among later backers. Those figures do not reconcile cleanly: a $200 million seed and a $235 million Series A sum to $435 million, while $550 million is the widely reported total, and the valuation and the Nvidia investment come from press reporting rather than from the company. The product is described as an operating system for science. Advanced models generate hypotheses, proprietary instruments the company calls AI Science Factories design and run the experiments, and results feed back in real time. Scientific Superintelligence and AI Science Factory are both presented as trademarks. Commercially there are two routes, neither specific to an industry: Catalyst applies the platform as an operating layer on a partner's existing research programmes, and Creation is a longer horizon partnership intended to originate discoveries, launch products and found new companies. Seven industries are given equal billing on the company's own navigation: Advanced Materials, Therapeutics, Chemicals, Oil and Gas, Energy and Environment, Aerospace and Defense, and Biotech. This record exists under a ruling that a vendor treating healthcare as a major vertical with a major healthcare offering is admissible even while selling elsewhere. Two of the seven are life science, and the therapeutic scope claimed is the widest in this lane, covering messenger RNA, proteins, antibodies, cell therapies and small molecules. The therapeutic claims are specific and entirely self reported. The company states its platform has produced thousands of discoveries, large language models with state of the art scientific reasoning, genetic medicine constructs performing better than commercially available therapeutics, and hundreds of antibodies, peptides and binders. On 28 August 2026 it published a blog stating that the platform first discovered an ultra stable messenger RNA and then produced a CAR-T therapy that outperformed a leading therapy used in patients today, in non human primates, designed by three scientists in a matter of months. As of 29 August 2026 no paper, preprint or dataset supporting any of this was located, and the account circulated through the company blog and social media rather than through a scientific venue. Unpublished primate data is not unusual at this stage in this field, but the size of the claim and the absence of anything checkable behind it should be read together. The checkable third party relationships are elsewhere. In July 2026 Lila joined the United States Department of Energy Genesis Mission with Phase I awards for three projects alongside Caltech, Lawrence Berkeley National Laboratory, Northwestern and Argonne, which is a competitive federal selection rather than a self assessment. A June 2026 collaboration with Nvidia covers the BioNeMo agent toolkit, though Nvidia is also reported as an investor, so that relationship is not arm's length.

Chemify is the only company in this lane that makes the molecules. Everything else here designs and hands the design to somebody with a laboratory. Chemify built the laboratory and is trying to turn it into a printer. The company spun out of the University of Glasgow, founded by Professor Lee Cronin, who holds the Regius Chair of Chemistry there and continues to run a laboratory at the university. Its own About page dates the founding to 2022; its own 2023 funding announcement dated it to 2019, and third party accounts reconcile the two as research beginning in 2019 with formal incorporation in 2022. The 2022 date is recorded here as the current first party statement. Headquarters are at the Advanced Research Centre in Glasgow, with the Maryhill Chemifarm a short distance away. Headcount passed 200, over half holding advanced chemistry or engineering degrees, which is the largest workforce graded in this lane. The technology is called Chemputation, and what it is depends on who is describing it. The company calls it a universal chemical programming language combined with advanced robotics and artificial intelligence, and in its own press releases calls Chemputation itself its purpose built artificial intelligence. A chemistry society describing the same system calls it a chemistry focused programming language, robotics and a database of reactions. Those are not the same claim, and nothing published reconciles them. What is not in dispute is the ambition: code goes in, physical matter comes out, and the reaction library the company says is the world's largest constrains what can be made to what has been validated. Chemifarm opened in Maryhill in June 2025 at a stated cost of $16 million, described by the company as the world's most advanced automated synthesis facility. A Silicon Valley site is planned. Funding runs to roughly $107 million across a $43 million Series A in August 2023 led by Triatomic Capital, which included $9 million from the United Kingdom Innovation Accelerators programme, and a Series B above $50 million in October 2025, with Insight Partners, Wing Capital, Founders Fund, DCVC, 8VC, BlueYard and Horizon Ventures among the backers. The leadership composition is unusual for a drug discovery company and tells you what this one thinks it is. The chief operating officer came from Virgin Media O2, Amazon and Tesco supply chain roles. The chief technology officer came from automotive and safety critical systems engineering. The advisory board includes a former chief supply chain officer of 3M, a former chief medical officer of Eli Lilly, and the global head of artificial intelligence and machine learning at GSK. This is a manufacturing leadership team wrapped around a chemistry platform. Commercially the disclosed relationships are with the Gates Foundation and Lgenia rather than with pharmaceutical companies. Gates funding for tuberculosis and malaria work reached $3.2 million cumulative after a $1.6 million follow on grant in December 2025, with Lgenia running enzyme and whole cell assays against compounds Chemify designs and synthesises. A malaria programme was described in December 2025 as moving toward investigational new drug enablement within twelve months and aiming at a single dose cure. As of 29 August 2026 no pharmaceutical company partnership, commercial contract or licensing agreement appears anywhere in the public record, which for $107 million raised and a facility in operation is the notable gap. One provenance point for anyone weighing the science. In April 2026 three peer reviewed papers appeared in PNAS, Nature Communications Chemistry and Nature Communications Biology demonstrating Chemputation for molecular discovery, automated synthesis from published literature, and cancer drug work. They were authored by Cronin with co authors from the University of Glasgow, where he holds his chair. The work is peer reviewed. It is not independent of the company's founder.

Deep Genomics has the deepest scientific record in this lane and the thinnest commercial one, and the gap between those two halves is the most useful thing about this entry. Brendan Frey launched the company in Toronto in 2015 out of his University of Toronto work on the genetic determinants of disease; he also cofounded the Vector Institute for Artificial Intelligence. Corporate headquarters sit on University Avenue in Toronto with a second office at Canal Park in Cambridge, Massachusetts. Brian O'Callaghan, a career pharmaceutical executive with prior chief executive roles at ObsEva, Petra Pharma, Acucela, Sangart and BioPartners, became chief executive in September 2023, with Frey moving to chief innovation officer and remaining on the board. Thong Q. Le and Paul Sekhri joined the board in October 2025. The platform is a set of biological foundation models rather than a single system. BigRNA carries close to two billion tunable parameters and was trained on over a trillion signals drawn from high throughput sequencing, learning from paired genotype and RNA expression data across many individuals; it predicts tissue specific regulation, protein and microRNA binding sites, and the effects of variants and candidate therapeutics. REPRESS predicts cell type specific microRNA binding and messenger RNA degradation directly from sequence. DeepADAR designs guide molecules for RNA editing, trained on sixteen million endogenous editing sites and fine tuned on a custom synthetic screening dataset. The company runs lab in the loop workflows across two facilities totalling more than 10,000 square feet, validating predictions with orthogonal and functional assays and feeding results back into training. The publication list is the longest and most prestigious graded in this lane, running to seventeen items including work in Science, Nature Biotechnology twice, Nature Medicine, Molecular Systems Biology and Proceedings of the IEEE. A reader should separate two things inside it. The landmark peer reviewed papers are largely Frey's academic lineage from 2015 to 2018, published around or before the company's product era. The work describing the actual commercial models, BigRNA, REPRESS, FlashRNA and Enigma, sits entirely on a preprint server. The BigRNA manuscript has been public since September 2023 and no peer reviewed publication of it was located as of 29 August 2026, which is close to three years for the company's flagship claim. The commercial picture is where this record diverges sharply from its neighbours. In September 2023 the company announced a strategic plan to evolve into a forward integrated biopharmaceutical company advancing multiple RNA targeted therapeutics. As of 29 August 2026 the site presents no pipeline, no therapeutics section and no named indication, and describes itself instead as an artificial intelligence foundation model platform working in support of major pharmaceutical partners. Not one of those partners is named, and no collaboration, deal value or programme outcome appears anywhere on the site or in the news archive. Two dedicated passes located no announcement explaining the change in direction. Every other vendor graded in this cluster names at least three partners. One detail that is easy to miss and worth carrying. Yann LeCun, who supplied the endorsement quote in the BigRNA announcement, sits on the company's scientific advisory board, so that praise is a related party statement rather than independent commentary.

Antiverse is the small independent in the antibody design cluster, and it is chasing the same target class as Nabla Bio on roughly a tenth of the money. Antiverse Ltd is registered in England and Wales under company number 10736975 and works out of the sbarc | spark building on Maindy Road in Cardiff, a Cardiff University innovation site, with further offices in Boston and Prague. Murat Tunaboylu is chief executive and cofounder, alongside Ben Holland and Rowina Westermeier. The founding date is disputed: business databases and the company's own claim of training data purpose built since 2017 both point to 2017, while one trade report has the company founded in 2020 and formally launched in 2022. The earlier date is recorded here on the weight of the evidence. The focus is the hardest antibody problem there is, and it is the same one Nabla Bio picked: G protein coupled receptors, ion channels and other transmembrane targets, which resist conventional discovery because they will not hold their shape outside a cell membrane. Antiverse attacks it from a different direction. Rather than a single generative system, the company describes three assets working together: proprietary training data built specifically for this target class rather than for proteins in general, machine learning generated epitope specific libraries, and programmable cell line engineering, with more than 240 proprietary stable cell lines for these receptors. Designs are built and characterised in house, in what the company calls a lab in the loop, with prioritised clones said to hold their natural conformation so characterisation completes in about a month. Formats extend to any biologic carrying a heavy chain. The partnership record is real but thinly disclosed. Nxera Pharma, formerly Sosei Heptares, signed a multi target partnership and licensing agreement in November 2024 pairing Antiverse's libraries with Nxera's structural platform, with the first project aimed at antibodies that activate rather than block a difficult receptor; Antiverse takes an upfront payment, research funding and milestone eligibility while Nxera holds an exclusive worldwide licence to the resulting assets. A research agreement with the Cystic Fibrosis Foundation announced in March 2026 targets the external face of the CFTR protein. Work with the United States biotech GlobalBio produced candidates against PD-1 that entered preclinical development. Two further pharmaceutical partners are acknowledged only by size, one described as a top 15 company and the other as a top 30. Funding is modest and the record has one unresolved conflict. A $9.3 million Series A led by Soulmates Ventures closed in March 2026, bringing the stated total above $20 million since inception, following an earlier seed extension of £3.5 million. Against that, one trade report states the company launched in 2022 with an initial £50 million from United Kingdom government funds and the Cardiff Capital Region, a figure that cannot be reconciled with the company's own total and is more likely the size of a regional fund than money received. It is recorded here as unreconciled rather than as either fact or error. One observation that matters more for how this vendor is found than for what it does. Both the homepage and the legal pages carry a noindex directive, and the terms of service prohibit automated access by bots, scrapers or spiders without prior written consent. A company whose commercial future depends on pharmaceutical partners finding it has instructed crawlers to stay away.

BigHat Biosciences is the closest thing in this lane to a company that shows its work, and the contrast with its nearest neighbour is instructive. Nabla Bio designs antibodies from nothing. BigHat starts from an existing molecule, its own or a partner's, and iteratively transforms it. Both are machine learning antibody companies with their own laboratories, and they are solving different halves of the problem. The company was founded in San Mateo, California in 2019 by Mark DePristo and Peyton Greenside, who had worked together at Google, with Theresa Tribble as founding advisor. DePristo built the genomics team inside Google Brain and previously co directed medical and population genetics at the Broad Institute. Greenside was a Schmidt Science Fellow at Stanford. Leadership has since changed: Greenside is now chief executive and DePristo appears on the company's own site as an advisor. The leadership page also lists a general counsel and a chief information officer, which is unusual at roughly 80 people and matters for what follows. The platform is Milliner, a machine learning layer welded to a synthetic biology wet lab that synthesises, purifies and fully characterises hundreds of antibody variants a week in a single workcell. Every program runs as a design, build and test cycle, with the computational models continuously updated by laboratory feedback. Greenside has stated the methodological position plainly: the company works from small and medium data pools chosen for informativeness rather than from large ones, on the argument that fast feedback beats volume. Formats covered include multispecifics, single domain antibodies, single chain variable fragments and antibody drug conjugates. The publication record is the strongest in this lane and it is not close. A dedicated publications page lists eleven items running from 2021 to 2026, at machine learning venues rather than in the biology press: work at ICML in 2021 and 2022, ICLR in 2024, 2025 and 2026, and NeurIPS in 2025, on Bayesian optimisation for antibody design, denoising autoencoder surrogates, guided sequence and structure generative modelling, masked language models for protein engineering, and a unification of discrete, Gaussian and simplicial diffusion. The thread is coherent rather than scattered, and two of the company's advisors are among the researchers whose Bayesian optimisation work the papers build on. Venue labels are the company's own and at least one entry is a workshop track rather than a main conference, so a reader should check each rather than take the list as uniform. Commercially it has five named pharmaceutical partners: Amgen, Merck, Johnson & Johnson through Janssen Biotech in neuroscience, AbbVie, and Eli Lilly. The AbbVie agreement is the most specific financial disclosure located anywhere in this lane, at $30 million upfront against up to $325 million in milestones. The April 2025 Lilly collaboration covers up to two antibody programs, came with an equity investment, and brought support for BigHat's own gastrointestinal cancer programme through Lilly Catalyze360 while leaving BigHat full global rights. A second Lilly project in January 2026 has BigHat generating datasets to feed a generalisable antibody developability foundation model inside Lilly TuneLab. The company in licensed site specific antibody drug conjugate technology from Synaffix, a Lonza company, in November 2024, and has acquired Frugi Biotechnology. Two things to hold carefully. The pipeline claim is forward looking: the company states a gastrointestinal cancer conjugate entering the clinic in 2026 and a T cell engager with an investigational new drug application in 2026, but as of 29 August 2026 no clearance, trial registration or first dosing has been announced, and a third party pipeline database still places the lead programme at the enabling stage. And the site contradicts itself on scale, with an About page reporting more than $100 million raised while the company's own January 2026 announcement says more than $140 million. That About page also still carries placeholder Latin text in its awards section and links several awards to a legacy domain that is not the company's site.

Nabla Bio designs antibodies on a computer and then makes them, and it has published more about how well that works than anyone else graded in this lane. The company is a Harvard spinout from George Church's laboratory, founded by Surge Biswas, who is chief executive, and Frances Anastassacos, who is president. It launched publicly in December 2021 with $11 million from Khosla Ventures, Zetta Venture Partners and Fifty Years, and is based at the Riverside Technology Center on Memorial Drive in Cambridge, Massachusetts. Company material dates the launch to 2021 while some databases record incorporation in 2020; the first party date is used here. Total funding stands near $37 million after a $26 million Series A led by Radical Ventures in May 2024. The platform is JAM, for Joint Atomic Modeling, and the company describes it as molecular autocomplete: given a target and a binding site, it completes the antibody. Two modules do the work, a Generator that builds complete protein complexes from partially specified ones, and a Filter that scores how likely a complex is to survive an experiment, with output fed back for iterative refinement across rounds the company calls introspection. The focus is deliberately on the hardest class of target, multipass membrane proteins such as G protein coupled receptors, ion channels and transporters, where conventional discovery struggles because the target cannot easily be presented outside a cell membrane. A successor system, JAM-2, was announced with an openly published technical report. The evidence is unusually specific for a private company. A January 2025 preprint reported the first fully computationally designed antibodies against multipass membrane proteins, Claudin-4 and CXCR7, at double digit nanomolar affinity in single domain and paired formats. A May 2025 preprint applied test time scaling, letting the model compute longer at generation rather than retraining it, and reported hundreds of designs against CXCR4 and CXCR7 reaching picomolar to low nanomolar affinity, including what the company says are the first antibody agonists reported for CXCR7 and the first computationally designed antibody agonists of any kind. Both preprints publish bind rates per round rather than only successes. Neither has been certified by peer review, and both carry the declaration that all contributors are current or former employees holding equity. The commercial record matches. Collaborations with AstraZeneca, Bristol Myers Squibb and Takeda were announced together in May 2024 at more than $550 million in combined upfront and milestone payments plus royalties, and Takeda, which had worked with the company since 2022, signed a second and larger agreement in October 2025 for double digit millions upfront against roughly $1 billion in potential value, applying JAM across its early stage programs. The company describes itself as revenue generating. One detail belongs on the record because it is rare. In 2024 an AstraZeneca scientist publicly said the generative outputs were fairly low affinity at that point, echoing broader criticism of de novo antibody work. A partner qualifying the technology in the press while continuing to fund it is a more useful signal than any case study, and the affinities reported a year later are the company's answer to it.

Aqemia is the physics first entry in this lane, and its central claim is a negative one. It says it does not need target specific experimental data to begin. The company was spun out of the École normale supérieure in Paris in 2019 by Maximilien Levesque and Emmanuelle Martiano-Rolland, building on more than a decade of research into quantum and statistical mechanics. It is headquartered on Boulevard Pasteur in Paris, with a second site on Euston Road in London opened at the end of 2024. Levesque is chief executive and Martiano-Rolland is chief operating officer. Reported headcount runs between roughly 78 and 200 depending on the source and the date. The platform is Qemi. Statistical and quantum mechanics algorithms compute binding free energy and generate their own data, which then feeds a generative model that proposes molecules. Most of this lane trains on existing assay measurements and is therefore bounded by what has already been measured; Aqemia starts from calculation instead. Levesque's own analogy is a system with no literature to read but millions of teachers correcting it. If the claim holds, it is the reason the company can take on targets carrying little or no prior chemical data, which is what Sanofi says it has handed over. The commercial record is the strongest evidence available, and in one respect it is unusually clean. Servier did not simply sign. It ran a blind test of the technology in early 2021, then announced a formal immuno oncology collaboration in December 2021, renewed it in 2023 against a target it described as undruggable, and in June 2024 reported that a series of designed molecules had proven active in both in vitro and in vivo assays. Sanofi began with SARS-CoV-2 work in 2020, signed a multi year discovery agreement in December 2023 making Aqemia eligible for up to $140 million in upfront and milestone payments across programs, and expanded it in July 2026 by nominating a new therapeutic target and triggering a further payment. Janssen has also been named as a partner. Under the Sanofi arrangement Aqemia designs and Sanofi leads wet lab research, development and commercialisation. Two qualifications belong with all of that. The $140 million figure is a ceiling across all programs inclusive of milestones rather than money received, and the individual payments have never been separated out; the same headline is quoted by the chief executive in a French university interview as €140 million, so the currency of the original agreement is not settled on the public record. And nothing about the technology has been published in the peer reviewed literature. Two dedicated passes on 29 August 2026 found no paper, no benchmark and no disclosed accuracy figure, which means every claim about how well the physics engine performs traces back to the company or to a partner announcement written with it. Funding mixes equity and public money: a €30 million Series A in 2022 led by Eurazeo with Bpifrance and Elaia, a €30 million extension led by Wendel Growth taking the all equity round to €60 million, and a $7.4 million France 2030 award in March 2025 to extend the platform to RNA targets and epitranscriptomics. Aggregator totals running above $100 million do not reconcile cleanly with the disclosed rounds.

PostEra is the medicinal chemistry counterpart to the biology platforms that dominate this lane. Where Immunai, Cellarity and Enveda build models of cells, tissue or natural product chemistry to decide what to make, PostEra builds models of how to make it. The company was founded in October 2019 by Alpha Lee, Aaron Morris and Matthew Robinson, on research from Lee's laboratory at the University of Cambridge. It closed its United Kingdom operation in early 2020, relaunched as a United States entity, and joined the Y Combinator winter cohort. Morris is chief executive, Lee is chief scientific officer and Robinson is chief technology officer. The company's own privacy policy and its press releases both place it in Boston, Massachusetts, while third party databases list a Cambridge, Massachusetts address; the first party statement is recorded here. The platform is Proton, described as an end to end machine learning system that closes the design, make and test cycle of medicinal chemistry. Design combines proprietary data with chemistry foundation models to produce molecules satisfying several competing property constraints at once. Make is the part that separates this vendor from the rest of the field: every designed molecule carries a synthetic route, and routes are selected so that molecules can be made in parallel from shared intermediates. Test uses active learning to choose the experiments that return the most information per cycle. Manifold, the retrosynthesis component, is licensed as software with a public API, is documented in a public help centre, and is integrated into Optibrium's StarDrop. The commercial model is biopharma partnership rather than software subscription. The company states it has closed over $1 billion in artificial intelligence partnerships, including four multi year agreements with Pfizer and Amgen. The Pfizer relationship began with generative chemistry in December 2020, expanded into a $260 million AI Lab in January 2022, and expanded again in January 2025 to a stated $610 million with an antibody drug conjugate collaboration added, carrying $12 million upfront plus milestone and royalty eligibility. Those headline totals are deal ceilings including milestones rather than money received, and should not be read as revenue. PostEra also leads an antiviral drug discovery centre for pandemic preparedness funded by the National Institutes of Health, and is advancing an internal pipeline that has moved toward reproductive medicine. One caution for anyone researching this vendor. A separate company trading as Manifold at manifold.ai sells an unrelated life sciences platform and publishes a trust centre carrying SOC 2 Type II, NIST 800-171, HIPAA, TX-RAMP and GDPR compliance. None of those attach to PostEra. The names collide and the certifications are easy to misassign.

Immunai occupies a different position from most of this lane. It is not developing its own drugs. It sells pharmaceutical companies an understanding of the immune system, and it has been paid repeatedly for it. The platform has three named parts. AMICA, the Annotated Multi omic Immune Cell Atlas, is a proprietary clinically annotated single cell immunology database. AMICA-OS is the operating system layer combining that database with foundation models. The Immunodynamics Engine is the model of immune function itself. The data underneath spans single cell RNA, surface proteins, immune receptor repertoires and spatial gene expression, drawn from clinical and laboratory samples, and the stated uses are biomarker discovery, patient stratification, mechanism of action analysis and dose optimisation in drug development. The validation here is commercial rather than published, and it is unusually strong of its type. AstraZeneca began working with Immunai in late 2022 and has expanded three times: across oncology clinical programmes, then into inflammatory bowel disease in October 2025 in a deal worth up to $85 million for exclusive rights to a target Immunai had identified through the atlas, then again in May 2026 for up to $37.5 million across 2026 and 2027. Bristol Myers Squibb signed a multi year agreement in January 2026 and Boehringer Ingelheim followed in June 2026 for T cell target discovery. A partnership with the Parker Institute for Cancer Immunotherapy assembled what both parties describe as the largest single cell dataset for real world immunotherapy research, from 3,700 blood samples across 1,070 patients treated with checkpoint inhibitors. A sophisticated buyer returning three times, and paying for a target the model found, is a harder signal to manufacture than a case study. The company is headquartered in New York with offices in Tel Aviv, Prague and Zurich, employs more than 170 people, is led by chief executive Noam Solomon and has raised close to $270 million. What distinguishes this record within the lane is that the privacy and security axes genuinely apply. The chemistry led companies here train on molecules; this platform is built from patient samples with clinical annotation, and the atlas is enriched by work done for one partner and then used to serve others. That accumulation is the product's central advantage and also the question nobody has answered publicly.

Terray Therapeutics is a Los Angeles small molecule company whose founding argument is about data rather than models. Its chief executive Jacob Berlin put it plainly at the Series A: even the most powerful artificial intelligence cannot get past insufficient or unclear data. Everything else follows from that position. The company built an ultra high throughput experimental platform using dense microarray chemistry to generate its own measurements rather than train on what already exists. It reports quantitatively measuring more than five billion target ligand interactions in three years, which it characterises as roughly fifty times the entirety of publicly available chemistry data and states is doubling annually. The integrated platform combining that experimental engine with the modelling layer is called tNova, and it powers both internal programmes and partnered work. What separates this record from the rest of the lane is that the model is published. COATI, the company's chemistry foundation model, appeared in a peer reviewed journal with a digital object identifier rather than in a press release. The paper describes a pre trained multi modal encoder decoder of druglike chemical space built by contrastive learning from text and three dimensional representations of molecules, states the properties claimed for it, and demonstrates a generative optimisation method against a named model protein on an explicit multi parameter task of potency, solubility and druglikeness. The company also says it brought latent diffusion to small molecule design, and describes its tools as co pilots its medicinal chemists use daily, which is a more modest and more checkable autonomy claim than most here make. The internal pipeline is immunology. Partnered programmes run with Bristol Myers Squibb and with Calico, the Alphabet subsidiary, which is meaningful third party validation of the platform. Equity funding exceeds $200 million across a seed round, a $60 million Series A led by Madrona in 2022 and a $120 million Series B in October 2024 led by Bedford Ridge Capital and NVentures. Two things a reader should weigh. NVentures is NVIDIA's venture arm, and the same Series B announcement carries an endorsement of Terray's models from NVIDIA's vice president of healthcare, so the most prominent technical praise in the company's materials comes from an investor rather than from an independent party. And the company told the trade press in October 2024 that it expected its first compound in the clinic by 2026. No investigational new drug application or clinical trial start was located as of this assessment, which does not mean it has not happened but does mean the guidance is so far unconfirmed.

Cellarity is a Flagship Pioneering company in Somerville, Massachusetts, publicly unveiled in 2019 after incubation in Flagship Labs, that designs drugs against the behaviour of a whole cell rather than against a molecular target. It now describes itself as clinical stage and as developing cell state correcting therapies through integrated multi omics and artificial intelligence modelling. The founding argument is a critique of how the industry works. Target centric discovery reduces a disease to a single protein and builds a molecule to hit it, and the company's position is that assumptions which hold in a dish or an animal frequently fail to translate into humans, which is a substantial part of why drugs fail in clinical development. Its alternative uses single cell technologies and machine learning to build a quantitative picture of the network state that defines how a cell behaves, then searches for compounds that move a diseased cell back toward a healthy behaviour. Mechanism is characterised afterwards, and may turn out to involve known targets, novel ones, or several at once. That inversion is the most consequential thing about the company and it cuts both ways. It is a genuine attempt at the hardest problem in the field, and it means a molecule can be selected before anyone knows what it does. The usual interpretability check available to a medicinal chemist, whether the compound hits the target it was designed for, does not exist here by construction, and nothing published describes how mechanism is established to a standard regulators will accept or how off target liability is assessed when the programme did not begin with a target. The pipeline has reached the point where the thesis becomes testable. CLY-124, a first in class oral globin switching candidate for sickle cell disease, entered a first in human study in June 2025 in healthy volunteers and people with the disease, with 28 day data guided for the end of 2026. A myelofibrosis programme has in vivo preclinical data against standard of care expected in 2026 and a development candidate nomination targeted for the fourth quarter. A programme in metabolic dysfunction associated steatohepatitis runs in partnership with Novo Nordisk, and a collaboration in inflammatory bowel disease is also stated. Hematology and immunology are named as the first therapeutic areas. Ted Myles is chief executive. Investors include Flagship Pioneering, Baupost Group, BlackRock, Pictet, Hanwha Impact and Kyowa Kirin. One founding date discrepancy is recorded rather than resolved: Flagship's own company page states the business was founded in Flagship Labs in 2017, while the company's current releases state 2019. The later date is used here as the company's own present statement, with incubation beginning earlier.

Enveda is a Boulder, Colorado biotechnology company founded in 2019 by chief executive Viswa Colluru that applies machine learning to natural product chemistry, an approach it describes as non genomics based discovery and which sits apart from the protein structure and generative chemistry work that dominates the rest of this lane. The premise is a stated gap in the addressable chemical space. The company argues that a large share of historic pharmaceutical breakthroughs derive from a fraction of a percent of nature's chemistry, and that the obstacle has never been the value of that chemistry but the difficulty of working with it: identifying which molecule in a complex extract is active, resolving its structure, deciding whether it is amenable to medicinal chemistry, and obtaining material at scale. Its platform pairs metabolomics and mass spectrometry with machine learning to attack those steps, and it has announced a foundation model for metabolomics trained on what it describes as the world's largest collection of experimental mass spectra. It reports characterising the structure and function of more than one million natural compounds in four years. The output is a therapeutics pipeline rather than software. Ten development candidates are claimed across immunology and inflammation, obesity, fibrosis and neurosensory indications. The lead programme, ENV-294, is an oral agent for atopic dermatitis that entered Phase I in late 2024 following clearance of its investigational new drug application, which makes this one of the minority of platform companies in this lane with an asset actually in human trials rather than a platform and a promise. Funding reached $360 million with a $130 million Series C in November 2024 led by Kinnevik and FPV, with Baillie Gifford, Premji Invest, Lux Capital and Dimension Capital participating. A third party profile reports a higher cumulative figure since, which was not verified here. One governance question is specific to this platform and is not addressed anywhere published. A business built on screening the chemistry of plants and other organisms collected globally sits directly on top of the international regime governing access to genetic resources and the sharing of benefits with source countries. The Nature Conservancy is among the company's investors, which suggests the question has been considered internally. Nothing states how source material is obtained, what consent or benefit sharing arrangements attach to it, or how provenance is tracked into the training data. For a company whose central asset is a dataset derived from the natural world, that is the disclosure a careful reader would want.

Absci designs antibodies with generative models and then puts them into people, which is what separates it from most of this lane. Listed on Nasdaq as ABSI, headquartered in Vancouver, Washington, with an artificial intelligence research laboratory in New York and an innovation centre in Zug, it describes itself as a data first generative artificial intelligence drug creation company. The Integrated Drug Creation platform is a closed loop: generative models design candidates, a synthetic biology data engine tests them in the wet laboratory at a stated throughput of billions of cells a week, and the resulting data retrains the models. The company states it can go from an artificial intelligence designed antibody to a wet laboratory validated candidate in as little as six weeks. The design step is specific rather than general: given a target structure, the de novo model is used to designate a particular epitope, so antibodies are engineered against a chosen site rather than selected from what a screen happens to find, and separate optimisation models then tune the candidate for developability, half life and immunogenicity risk. Three programmes have reached the clinic. ABS-101, an anti-TL1A antibody for inflammatory bowel disease, was de novo designed on the platform, entered investigational new drug enabling studies in February 2024 and has reported interim Phase 1 data indicating an extended half life against first generation competitors in the class and a favourable safety profile; the company has said it is exploring partnership and out licensing rather than running later trials itself. ABS-201, an anti prolactin receptor antibody for androgenetic alopecia, reported positive interim Phase 1 safety, pharmacokinetic and immunogenicity data in June 2026 from an Australian trial of up to 227 participants, with proof of concept data expected from late 2026 into early 2027, and is also being explored in endometriosis. Preclinical work includes ABS-301, from a reverse immunology approach against an undisclosed immuno oncology target, and ABS-501, an anti-HER2 candidate reported to show activity in trastuzumab resistant models.

Deciphex is a Dublin company founded in 2017 by Donal O'Shea, its chief executive, and Mark Gregson, addressing the shortage of pathologists rather than the accuracy of any single diagnosis. Its chief medical officer is Runjan Chetty. It has raised roughly 56 million dollars in total, including a 31 million euro Series C in January 2025 led by Molten Ventures with ACT Venture Capital, Seroba, Charles River Laboratories, IRRUS Investments, the HBAN Medtech Syndicate and Nextsteps Capital. The business has two halves and a buyer should understand which one they are purchasing. Patholytix is software: a non clinical workflow platform for toxicologic and preclinical pathology, used by pharmaceutical and biotechnology companies during drug safety assessment. In April 2024 Charles River Laboratories, the largest preclinical research organisation in the sector and also an investor here, launched Patholytix Foresight jointly with Deciphex, a decision support tool built on Patholytix 4.0 that pairs artificial intelligence classifiers with whole slide images to speed primary evaluation and peer review. The two extended that into an exclusive image management arrangement in February 2025. Diagnexia is the other half and it is a service rather than a product: a network of more than 250 subspecialty pathologists who report cases digitally for healthcare providers, positioned explicitly against the traditional locum model, with a research variant called Diagnexia Analytix serving drug development. This record is scoped principally to the software, since Patholytix is separately licensable while Diagnexia is a staffed diagnostic service, and both are described here because the artificial intelligence and the human network are sold together. The company states its platforms let pathologists work up to 40 percent faster while maintaining accuracy, is expanding across the United States, United Kingdom, European Union, Canada and Japan, holds a partnership with Novartis on artificial intelligence for drug discovery pathology, and has stated it will use its image repository to build pathology foundation models.

Aignostics builds foundation models for computational pathology and sells their output to biopharmaceutical companies for drug discovery, translational research, clinical trials and companion diagnostic development. It was established in 2018 inside Charité Universitätsmedizin Berlin and the Berlin Institute of Health, alongside TU Berlin and Fraunhofer HHI, and spun out in 2020. Founders include Frederick Klauschen of Charité, Viktor Matyas and Maximilian Alber. It is based in Berlin with a New York presence and more than 75 staff. The models are the product and they are documented in public preprints rather than in marketing copy. RudolfV, the first, was built on a deliberately heterogeneous dataset drawn from more than 15 laboratories covering 58 tissue types and 129 histochemical and immunohistochemical staining modalities, with pathologist knowledge built into the curation rather than applied afterwards. Atlas followed, developed with Mayo Clinic and Charité, a vision transformer of roughly 632 million parameters trained on 1.2 million whole slide images from more than 490,000 cases, sampled into about 520 million tiles at four magnifications, and evaluated against 21 public benchmarks alongside named rival models. Atlas 2, announced January 2026 with Mayo Clinic, LMU Munich and Charité, is around 2 billion parameters trained on more than 5 million slide images and reports the highest average performance across 80 public benchmarks, with distilled smaller versions released for compute constrained settings. The most recent product, Atlas H&E-TME, is a self service application profiling the tumour microenvironment at single cell resolution from routine stained images. The company states that Atlas 2 ships with clinical grade regulatory documentation intended to support integration into medical devices built by others. More than 55 million dollars has been raised including a 34 million dollar Series B in October 2024, with ATHOS, Wellington Partners and the Boehringer Ingelheim Venture Fund among investors. Development partnerships are named with Bayer and Mayo Clinic.

Protein design company built on the ProGen family of protein language models, originating in a Salesforce research project that first demonstrated large language models could generate functional proteins, published in Nature Biotechnology in 2023. ProGen3 is a family of billion parameter generative models trained on more than 3.4 billion protein sequences using a sparse architecture the company reports delivers a fourfold speedup, and it generates full length novel proteins or redesigns specific domains of an existing protein. The company's defining result is OpenCRISPR-1, described as the first CRISPR gene editor designed from scratch by AI and published in Nature in 2025: from generated candidates, 48 sequences were functionally characterised in human cells, and the lead showed comparable on target editing to SpCas9 at 55.7 percent against 48.3 percent while cutting off target editing by roughly 95 percent, at 0.32 percent against 6.1 percent, sitting 403 mutations from SpCas9 and 182 from any natural protein in the CRISPR-Cas Atlas. OpenCRISPR-1 was released for free licensing, which also sidesteps the licence payments attached to existing CRISPR patent families, and the company reports tens of thousands of downloads within a day and use across academic, pharmaceutical and commercial operations. Further work includes E1, described as the first retrieval augmented model for protein engineering, Protein2PAM for programming the DNA motifs an editor recognises, and OpenAntibodies covering single shot antibody design against 20 drug targets. Commercial paths are asset licensing, collaboration, or early access to the API, with named relationships including Revvity, Corteva Agriscience, the Rett Syndrome Research Trust and Integrated DNA Technologies. Total funding is 150 million dollars including a 106 million dollar Series B co led by Altimeter Capital and Bezos Expeditions.

Clinical stage precision medicine company (Nasdaq: RLAY) built around the Dynamo platform and a thesis it calls motion based drug design: that proteins move, and that binding sites appear only in particular conformational states, so a discovery approach anchored to static structures misses targets that a dynamic one can reach. Dynamo integrates molecular dynamics simulation with experimental structural biology and machine learning across three phases, beginning with synthesising full length proteins through in house protein engineering to characterise the conformational dynamics of all domains of a target and form a modulation hypothesis. Screening data from activity based, ligand centric and computational screens feeds the machine learning components, including a proprietary machine learning powered DNA encoded library capability the company terms REL-DEL, strengthened by its acquisition of ZebiAI. The company reports that the integration of computation and experiment yields a larger number of chemical series entering lead optimisation. The disclosed focus is small molecule precision oncology with additional reported work in fibrosis and genetic disease, and the company has multiple programmes in clinical trials, with RLY-2608, a mutant selective PI3K alpha inhibitor for breast cancer, reported as its most advanced. A partnership with X-Chem announced December 2023 applies DNA encoded library screening alongside Dynamo against a GPCR target.

Drug discovery company built around super resolution live cell fluorescence microscopy, applying advanced optics descended from Nobel recognised single molecule imaging work together with machine learning to track how individual proteins behave and move inside living cells, rather than inferring behaviour from fixed or lysed samples. The company was founded in 2019, emerged from stealth in 2021 under chief executive Roger Perlmutter, formerly head of research and development at Merck, and has raised more than 1 billion dollars in venture capital including a 517 million dollar Series B in 2022 and a Series D of roughly 351 million dollars, before listing on Nasdaq under the ticker EIKN in a 2026 offering reported to have raised 381 million dollars. It reports a pipeline of more than 15 in house drug programmes built on the platform. An important scoping distinction for buyers: the assets currently in clinical trials were not generated by the platform. The lead candidate EIK-1001, a TLR7/8 dual agonist in Phase 2/3 trials for advanced melanoma and stage IV non small cell lung cancer in combination with pembrolizumab, was acquired from a partner, and two PARP inhibitors in early stage trials, EIK-1003 and EIK-1004, were licensed from Impact Therapeutics. The platform derived programmes remain preclinical as of this review.

AI research lab building protein language models, founded by researchers who emerged from Meta's Fundamental AI Research unit where they had developed ESM1, the first transformer language model for proteins. Its flagship model ESM3 reasons simultaneously over the sequence, structure and function of proteins in a single all to all architecture, meaning structure and function annotations can be supplied as inputs rather than only produced as outputs, which lets a scientist prompt the model toward a protein with specified properties. ESM3 was trained on 2.78 billion protein sequences at 98 billion parameters using more compute than any previously disclosed biological model. Its headline demonstration was esmGFP, a novel green fluorescent protein distant enough from natural variants that the company describes it as equivalent to simulating hundreds of millions of years of evolution. ESM3 was published in Science in January 2025. The model family comes in three sizes reached through the company's Forge API and partner platforms including AWS SageMaker and NVIDIA BioNeMo as an NIM microservice, while ESM3-open, a deliberately smaller model, has weights and source code on GitHub under a non commercial licence. The company states all models are built and deployed under a responsible development framework, and the Science paper records an external review of the risks and benefits of releasing ESM3-open. Raised 142 million dollars in seed funding led by Lux Capital, Nat Friedman and Daniel Gross with NVIDIA and Amazon participating.

Stanford spinout building GEMS (Genesis Exploration of Molecular Space), a small molecule discovery platform that integrates language models, diffusion models and physics based machine learning simulations to generate novel molecules and predict properties including potency, selectivity, ADMET and pharmacokinetics. The stated focus is targets that are biologically well validated but considered undruggable because the chemistry is difficult, including modelling how molecules bind to flexible or otherwise difficult proteins. The company runs both an internal pipeline and partner programmes, staffed by forward deployed engineers and drug hunters who work inside partner teams, with the stated intent that every partnered and internal programme stress tests GEMS. Disclosed collaborations: Genentech; Eli Lilly, reported at up to 670 million dollars with 20 million upfront; Gilead, 35 million dollars upfront across three targets with an option to nominate more at a predetermined per target fee; and Incyte, initiated 2025 and expanded 2026, with 150 million dollars in total upfront consideration including a 40 million dollar equity investment. The 2026 Incyte expansion also involves Incyte sharing significant experimental data for use in training GEMS. Total capital raised exceeds 280 million dollars, including a 200 million dollar Series B co led by Andreessen Horowitz with Fidelity, BlackRock and NVIDIA's venture arm participating. The company renamed from Genesis Therapeutics to Genesis Molecular AI in late 2025, and its published email addresses still resolve to the former genesistherapeutics.ai domain while the site is at genesis.ml. In October 2025 it released Pearl, a generative diffusion model for protein and ligand structure prediction presented as a core component of GEMS and described as trained on physics based synthetic data proprietary to the company.

Amsterdam based protein engineering company selling an enterprise AI software platform directly to R&D teams rather than operating as a discovery partnership. The platform uses generative machine learning to design and optimize proteins for specified properties, effectively reverse engineering a sequence from a desired function, and covers 3D structure prediction, thermostability optimization and codon expression alongside design suggestions. Models are trained on the customer's own data and tailored to that customer's programmes, and Cradle operates its own wet lab to expand its foundational models across additional protein modalities and properties. Two commercial features are unusual for this category: customers retain full ownership of any proteins engineered on the platform, and the product is sold as software that a pharma scientist team uses directly rather than as a collaboration. As of the December 2025 update the platform served six of the top 25 global pharmaceutical companies across more than 50 R&D programmes, with named users including Johnson & Johnson, AbbVie, Novo Nordisk, Novonesis, Grifols and Lundbeck, and applications spanning therapeutics, diagnostics, agriculture, food production and chemical manufacturing. Total funding exceeds 100 million dollars, including a 73 million dollar Series B led by IVP with Index Ventures and Kindred Capital.

Drug and materials discovery platform (HKEX: 2228) founded in 2015 by three MIT physicists, combining quantum physics first principles calculations, generative AI and large scale robotic automation in a closed loop Design Make Test Analyze cycle. Named components include XFEP for binding affinity prediction and a multi agent system orchestrating automated chemical synthesis, with the robotic laboratory infrastructure intended to close the historical gap between computational design and wet lab validation. In February 2025 XtalPi became the first company to list under the Hong Kong Stock Exchange's Chapter 18C framework. The business spans two revenue lines, drug discovery services and intelligent robotics, and reaches over 300 enterprise and research clients across pharmaceuticals, agriculture, energy and cosmetics. Disclosed 2024 revenue was 266 million yuan, up 53 percent, of which the intelligent robot business contributed roughly 163 million yuan and drug discovery roughly 104 million yuan. Platform derived assets and deals include PEP08, a PRMT5 inhibitor discovered for PharmaEngine that received clinical clearance in June 2025 and began Phase I enrollment in solid tumors in Australia and Taiwan; a collaboration with DoveTree Medicines announced August 2025 worth up to 5.99 billion dollars in upfront and milestone payments; and a June 2026 partnership on an oral GPCR programme in metabolic disease with total potential value over 400 million dollars. Prior collaborations include Pfizer on molecular modelling and Eli Lilly on bispecific antibodies through the Ailux biologics subsidiary.

London based techbio company built around the Benevolent Platform and a biomedical Knowledge Graph, a network of contextualized scientific data covering genes, proteins, diseases and compounds and the relationships between them, assembled from scientific literature, patents, genetics, chemistry and clinical trials and queried using natural language processing, machine learning and large language models to predict novel drug targets. The platform is disease agnostic by design. Its longest running commercial validation is a collaboration with AstraZeneca begun in 2019 in idiopathic pulmonary fibrosis and chronic kidney disease and expanded in January 2022 to heart failure and systemic lupus erythematosus, under which AstraZeneca scientists work alongside Benevolent scientists and independently validate candidate targets before selecting them for portfolio entry, with multiple targets selected across those four diseases. A separate strategic collaboration with Merck was signed in 2023. The platform also first identified baricitinib as a COVID-19 treatment, which subsequently received FDA approval. Internal assets have had mixed outcomes: BEN-2293 for atopic dermatitis was stopped after Phase IIa with the company stating it would not invest further, while BEN-8744 for ulcerative colitis entered Phase I. Following a December 2024 strategic overhaul with restructuring and layoffs, the company delisted from Euronext Amsterdam on 13 March 2025 via a merger into Osaka Holdings, a Luxembourg entity that then adopted the BenevolentAI name; the original listed entity ceased to exist and the business is now private, with stated runway into 2027 and a strategy of converting the platform into standalone modular product offerings and partnering earlier stage assets sooner.

Clinical stage generative biology company (Nasdaq: GENB) founded in 2018 by Flagship Pioneering and launched in 2020, designing protein therapeutics de novo rather than discovering them from natural scaffolds. Its core model, Chroma, generates novel protein structures and sequences conditioned on desired properties such as binding specificity, thermostability, size, symmetry and functional constraints, learning patterns from the Protein Data Bank in a way that generalizes across natural proteins. Chroma was published in Nature in November 2023 and released as open source software with model weights made freely accessible to academic researchers and non profit entities. The wider Generate Platform pairs generative and predictive models with what the company calls biohardware, including DNA assembly, protein production, multiplexed assay miniaturization and an in house cryogenic electron microscopy core, in a continuous generate build measure learn loop, organized into three modular capabilities: programmable binding, programmable function, and programmable composition and developability. The lead asset GB-0895 is a long acting anti TSLP monoclonal antibody engineered for potential every six month dosing, in Phase 3 for severe asthma with the first Phase 3 patient dosed 26 January 2026 and in Phase 1b for COPD; GB-4362 received FDA Fast Track designation on 23 January 2026. Disclosed collaborations include Amgen and Novartis. The company had 312 employees as of 31 December 2025 including 138 MDs and PhDs, and its cryo-EM core generated over 500 high resolution maps in 2025.

South San Francisco company founded in 2023 and launched publicly in April 2024 with more than 1 billion dollars in committed funding, the largest initial funding commitment in ARCH Venture Partners history, co led by ARCH and Foresite Labs with F-Prime, NEA, Sequoia Capital, Lux Capital and Lightspeed among the syndicate. The stated aim is end to end application of AI across drug discovery and development, combining AI methods research, large scale biological and clinical data generation, and therapeutic product development. Co founders include Marc Tessier-Lavigne, former Chief Scientific Officer of Genentech and former president of Stanford University, who serves as CEO; David Baker, the University of Washington protein design researcher awarded the 2024 Nobel Prize in Chemistry; Hetu Kamisetty, Chief Technology Officer; Vik Bajaj; and Robert Nelsen. The research team includes people who developed the RFdiffusion and RFantibody protein and antibody design models in the Baker laboratory, and the company absorbed technologies and personnel spun out of Illumina's functional genomics R&D effort along with a proteomics group from Interline Therapeutics. The board includes former FDA Commissioner Scott Gottlieb, Nobel laureate Carolyn Bertozzi and former Johnson & Johnson CEO Alex Gorsky. Headcount was reported at roughly 204 as of May 2026. In March 2026, after two largely silent years, the company made its first substantial platform disclosures: X-Cell, a virtual cell model of roughly 4.9 billion parameters that predicts cellular responses to genetic perturbation, released alongside X-Atlas/Orion, presented as the largest publicly available genome wide Perturb-seq dataset, and a preprint describing the model's development. The reported training corpus is 25.6 million perturbed single cell transcriptomes across sixteen biologically diverse contexts. The company has since described its therapeutic focus as inflammatory and immunological disease, with antibody therapeutics as the starting point. No clinical programme, named pipeline asset or named pharmaceutical partnership had been publicly disclosed as of this review.

Computational chemistry software company (Nasdaq: SDGR) whose physics based molecular simulation platform is licensed to pharmaceutical, biotechnology, industrial and academic organizations, alongside a proprietary therapeutics pipeline built on the same platform. Life science products include Maestro (molecular modeling interface), Glide (docking), FEP+ (free energy perturbation for binding affinity prediction), LiveDesign (enterprise informatics and collaborative design, with an ML module for training and deploying predictive models) and BioLuminate (biologics). A parallel materials science product line covers polymers, organic electronics, catalysis and formulation. The company's stated technical position is that machine learning is most useful paired with physics based methods rather than treated as a substitute for them, which distinguishes it from generative first platforms in this category. The platform is built on more than 30 years of R&D investment. Proprietary clinical programs include SGR-1505, a MALT1 inhibitor that received FDA Fast Track designation for relapsed or refractory Waldenström macroglobulinemia, and SGR-3515, a Wee1/Myt1 inhibitor. In January 2026 the company announced that Lilly TuneLab workflows would be made available inside LiveDesign, with LiveDesign as a priority interface for participating biotech companies.

Technology first drug discovery company (Nasdaq: RXRX) built around the Recursion OS, a platform pairing high throughput biological experimentation with machine learning on cellular imaging and multiomic data to map biology at scale. In November 2024 Recursion completed its combination with Exscientia, adding precision chemistry design and automated small molecule synthesis to what had been a phenomics heavy biology platform, creating an end to end discovery stack spanning target exploration through chemistry. The company operates BioHive-2, described at completion as the fastest supercomputer wholly owned and operated by any pharmaceutical company, built with NVIDIA technology. Data assets include over 60 petabytes generated in house or licensed from partners including Helix and Tempus. The commercial model is dual: an internal pipeline plus pharma partnerships carrying disclosed milestone potential of more than 20 billion dollars before royalties, with over 450 million dollars in upfront and realized milestone payments received to date. In May 2025, roughly six months after the merger closed, the company deprioritized three clinical stage programs, paused another and wound down a preclinical program, narrowing to six active development projects concentrated in oncology and rare disease.

AI drug design company spun out of Google DeepMind in 2021 and founded by Demis Hassabis, building on the AlphaFold structure prediction lineage. In February 2026 the company released IsoDDE, a unified drug design engine combining structure prediction, ligand binding, affinity prediction and antibody interaction modeling in a single pipeline. Reported performance is roughly double AlphaFold 3 accuracy on the hardest ligand binding cases where the target has under 20 percent sequence similarity to training data, and 2.3 times AlphaFold 3 on antibody antigen docking in the high fidelity regime. Unlike AlphaFold 1 through 3, IsoDDE is proprietary: no code, no weights, no public API and no peer reviewed publication, and Nature reported that the technical paper offers scant insight into how the results were achieved. Access is available only through pharma partnership. Named partners are Eli Lilly and Novartis, announced January 2024 with combined potential milestone value near 3 billion dollars, expanded with Novartis in February 2025, plus Johnson & Johnson. The company raised 600 million dollars in March 2025 led by Thrive Capital with GV and Alphabet participating. First in human trials for an AI designed candidate are targeted for end of 2026, a timeline that slipped from an earlier 2025 target.

Structure based drug discovery company built around AtomNet, a patented deep convolutional neural network trained on structure activity data and protein structures to predict small molecule bioactivity in three dimensions, applied as a virtual alternative to physical high throughput screening across an ultra large synthesizable compound library. The company's central evidence is the AIMS initiative, published in Scientific Reports in April 2024 as the largest and most diverse virtual high throughput screening campaign reported to date: AtomNet was applied to 318 targets through collaborations with more than 250 academic labs across 30 countries, and found structurally novel hits for 235 of the 318 targets evaluated. The paper explicitly addresses historical limitations of computational screening by reporting success on targets with no known binders, without high quality crystal structures, and without manual cherry picking of compounds, and reports that selected molecules were novel drug like scaffolds rather than modifications of known bioactives. Commercial posture is partnership led rather than independent clinical development, with disclosed collaborations including Sanofi, Hansoh Pharma and Eli Lilly. A lead internal candidate is an allosteric TYK2 inhibitor discovered using AtomNet.

French generative AI company for medicinal chemistry, incorporated October 2016, selling its technology both as software and as discovery collaboration. Three named products cover the design make test cycle: Makya, a generative AI platform for de novo design and multi parametric optimization that builds synthetic accessibility into generation by leveraging commercial building blocks and organic reactions rather than filtering for it afterwards; Spaya, an AI retrosynthesis platform that converts target compounds into commercially available starting materials and surfaces plausible synthetic routes; and Ilaka, an orchestration layer managing the workflow from raw material ordering through synthesis campaigns. The company also operates its own robotics platform for synthesis, purification, analysis and testing, and states this integrated approach shortens discovery to under two years. Makya is offered as SaaS, for on premise implementation, or inside a customer's own virtual private cloud, and can be driven through a graphical interface or as a Python package via Jupyter. Iktos reports validation through more than 50 academic and industrial collaborations, with named partners including Pfizer, Janssen, Merck KGaA, Servier, Sanofi, Ono, Teijin, Galapagos, Chiesi and UCB, and a strategic collaboration with Servier reported at potential total value over 1 billion euros.

Machine learning driven drug discovery and development company founded in 2018 by Daphne Koller, self described as the AI therapeutics company built on causal biology. The approach converges in house generated multimodal cellular data (induced pluripotent stem cells, genome editing, high content cellular phenotyping) with high content human cohort data, using machine learning to identify genetic drivers and prioritize targets rather than to generate molecules alone. The POSH platform was validated in Nature Communications in December 2025, with the company's stated finding that self supervised models trained on unbiased cellular morphology can reconstruct gene function and causal relationships without being told what to look for. In January 2026 insitro acquired CombinAbleAI and launched TherML, a modality agnostic design platform that uses a physics informed optimization engine pretrained on over 100,000 molecular dynamics surrogates for complex biologics including multispecific antibodies and T cell engagers, and proprietary Quantitative Adaptive Libraries to map chemical space for small molecules. Programs are concentrated in metabolic disease and neuroscience. The company has raised approximately 800 million dollars including approximately 150 million from non dilutive pharma partnerships, with named collaborations including Bristol Myers Squibb, Eli Lilly and Genomics England.

Computational pathology company that produced the first FDA authorized AI based digital pathology product, Paige Prostate, cleared via the De Novo pathway for cancer detection support. Subsequently pursued foundation model scale work in oncology imaging and received FDA Breakthrough Device designation for a pan cancer detection tool spanning common and rare variants across multiple tissue types. Acquired by Tempus in August 2025 for a reported 81.25 million dollars, making it now part of an indexed vendor rather than an independent competitor.

Spatial biology company applying computer vision and machine learning to tissue imaging, integrating high plex spatial proteomics, histopathology, and clinical data to identify predictive spatial biomarkers. Positioned primarily for pharmaceutical R&D, supporting patient stratification and trial enrichment for antibody drug conjugates, bispecifics, and immunotherapies, with an emerging diagnostics application. Reported as the first spatial AI tool used by pathologists for clinical trial patient selection tied directly to a drug development program.

Clinical stage biotechnology company applying machine learning to human tissue derived multi omic data through its CONVERGE platform, which the company describes as an all in human closed loop system that pairs one of the field's larger proprietary patient tissue databases with computational target prediction, validated in its own wet laboratories. Known for taking an AI discovered ALS candidate into human trials; following a Phase 1 efficacy miss the company refocused on the discovery platform and pharmaceutical partnerships.

Clinical stage company applying a physics informed AI discovery platform to small molecule drug design. Two named model families anchor the platform: NeuralPLexer for protein ligand complex structure prediction and Enchant, a multimodal transformer for predicting biochemical and cellular activity from small data sets. The lead asset IAM1363, a selective HER2 inhibitor discovered on the platform, is in a Phase 1/1b trial in advanced HER2 altered cancers.

Commercial spinout of the nonprofit research lab FutureHouse, building an autonomous AI research platform for scientific R&D. Its flagship agent Kosmos runs extended research campaigns that read literature, execute analysis code, generate hypotheses, and return fully cited reports. Buyers are biopharma and biotech R&D organizations rather than providers, and the platform is sold on a credit model with an academic free tier.

Tech bio company combining biological large language models, multimodal patient data, and agentic software. The Owkin K co pilot has two environments: K Navigator, an agentic research environment free to academic researchers that accelerates literature review across 26.5 million articles and 19 biomedical databases and explores spatial multiomic patient data, and K Pro, an enterprise co pilot that uses a single orchestrator to select and combine specialized biological AI skills across drug discovery and development. Both are powered by Owkin Zero, a fine tuned biological reasoning model. The company reports K Pro accelerating internal drug target identification from more than 12 months to roughly 3 months, validated through collaborations with AstraZeneca, Bristol Myers Squibb, and Sanofi, including a three year AstraZeneca licensing agreement to build biopharma agents. Owkin operates a group of entities spanning a biology foundation model (Bioptimus), diagnostics (Waiv), and a clinical stage drug program (Epkin); this record covers the software platform. Its CE-IVD marked pathology diagnostics, RlapsRisk BC and MSIntuit CRC, are indexed separately as Owkin Dx. Founded 2016 by Thomas Clozel, MD and Gilles Wainrib, PhD.

AI foundation models for molecular design, licensed as software to pharmaceutical and life sciences R&D organizations. The models predict and reprogram interactions between biochemical molecules: Chai-1 (2024) for structure prediction, Chai-2 (2025) for fully de novo antibody design, and Chai-3 (2026), which the company describes as roughly doubling the success rate of its predecessor. On Chai-2, the company reports designing all complementarity determining regions from a target and epitope prompt alone, with hit rates reported between roughly 16 and 20 percent depending on the source, against sub 0.1 percent rates the company attributes to prior methods; validation covered approximately 50 antibody targets with fewer than 20 designs tested per target. Models are reported in production at Eli Lilly, Pfizer, and Novartis, with a collaboration announced with argenx. Unlike vendors that use AI to build their own drug pipeline, Chai's product is the model itself, including custom versions trained on a customer's proprietary data.

Generative AI drug discovery company (HKEX: 3696) operating both as a platform vendor and a clinical stage biotech. Pharma.AI comprises PandaOmics (AI target identification and indication prioritization), Chemistry42 (generative molecular design using generative tensorial reinforcement learning rather than library screening), and inClinico (clinical trial outcome prediction). The platform's flagship validation is rentosertib, a TNIK inhibitor for idiopathic pulmonary fibrosis where AI selected the target, generated the molecule, and informed trial design: it entered a 320 patient Phase III trial in July 2026, with discovery published in Nature Biotechnology and Phase IIa results in Nature Medicine. The vendor reports reaching preclinical candidate nomination in 12 to 18 months on average against a 2.5 to 4 year industry norm, and 13 programs cleared for IND.