Enveda

PRISM, Enveda's foundation model for natural chemistry, identifies molecules in plants, microbes and the human body, predicts their structures from mass spectrometry data, guides laboratory experiments to reveal what they do and ranks them by therapeutic potential. It predicts a structure in about half a second, against the weeks a nuclear magnetic resonance experiment takes for a single molecule. The company says teaching AI the language of chemistry moved analysis from one molecule at a time to 10,000 at a time.

The library it reads was built by linking a database of 38,000 plants to 12,000 human diseases and symptoms, sourcing the plants with the highest likelihood of therapeutic potential and profiling them with mass spectrometry and machine learning. Since 2019 that approach has produced 17 development candidates, three of them now in human trials, and ENV-294 and ENV-308 were both discovered with PRISM. Enveda also applies its AI to its own manufacturing, quality and clinical trial operations.

Molecules move from prediction to development candidate through the laboratory and the company's chemists. PRISM predicts structures and ranks molecules by therapeutic potential, a fully automated laboratory screens them and determines their therapeutic properties at high throughput, and chemists refine the most promising into oral medicines, so an experimental result sits between each prediction and a candidate. Model outputs are presented as predictions and forecasts, and Enveda notes that only a few percent of molecules in a typical biological sample can be identified against known reference spectra.

No published rule says which model outputs a person must approve before a molecule advances.

PRISM, short for Pretrained Representations Informed by Spectral Masking, is a self supervised encoder adapted from the BERT family of natural language encoders. It learns by masked peak modeling: the masses of 20 percent of the peaks in a tandem mass spectrum are and the model predicts them from the context that remains. Training used 1.2 billion small molecule mass spectra and 85 billion tokens, roughly 600 million spectra from three public repositories, GNPS, MetaboLights and the Metabolomics Workbench, and roughly 600 million from Enveda's internal metabolomics platform.

The technical post describing the model, dated May 28, 2024, links earlier in house models such as MS2Mol and MS2Prop and cites outside methods including CSI:FingerID, MIST, MSNovelist and MS2DeepScore.

No version or release history is published.

Enveda gathered the training set and trained PRISM itself, on an architecture adapted from the published BERT family of language encoders rather than licensed from a model vendor. Microsoft is the one infrastructure supplier named, under a collaboration announced on May 29, 2024 that includes access to compute resources. Three public metabolomics repositories, GNPS, MetaboLights and the Metabolomics Workbench, supply about half of the 1.2 billion training spectra, and Enveda's own profiling platform supplies the rest.

No subprocessor or supplier list is published.

Three Enveda medicines are in human trials and two have reported results. ENV-294, a once daily oral agent for moderate to severe atopic dermatitis, reported Phase 1b results on March 31, 2026 from an open label study of nine adults dosed at 800 mg for 28 days with a 14 day off treatment period: a mean 68 percent reduction in Eczema Area and Severity Index at day 28 deepening to 85 percent at day 42, EASI-50 in all patients, EASI-75 in 78 percent and EASI-90 in 56 percent at day 42, 44 percent reaching an Investigator Global Assessment score of 0 or 1, and no serious or severe adverse events, treatment related adverse events or discontinuations. Randomized, double blind, placebo controlled Phase 2a studies in atopic dermatitis and asthma began in December 2025, and the atopic dermatitis study, NCT07298395, is recruiting an estimated 105 adults. New mechanistic and biomarker data from the Phase 1b study were accepted for a late breaking oral presentation at the EADV Congress 2026 in Vienna.

ENV-308 reported Phase 1 results on August 18, 2026 from 88 healthy volunteers, with no serious adverse events, discontinuations or dose interruptions and a fall in circulating leptin. ENV-6946 entered Phase 1 in December 2025.

For the platform, Enveda reports 75 percent fewer analogs to a development candidate, an 11 times success rate per scaffold and 4 times faster progress to a development candidate, each footnoted to a published industry figure (676 analogs per candidate, roughly 1 in 40 lead optimization scaffolds progressing beyond preclinical testing, 55 months to a candidate) and, for scaffolds, to its own rate of 1 in 4.

None of these clinical results has yet appeared in a peer reviewed journal.

PRISM is trained on small molecule mass spectra rather than on records about people. Personal health information enters through Enveda's trial recruitment pages: the atopic dermatitis page asks visitors their age, whether a doctor has diagnosed them and how severe their eczema is, the asthma page asks about diagnosis, daily controller inhaler use, recent biologic therapy and flare ups treated with systemic corticosteroids, and visitors who qualify leave a name, phone number and ZIP code and consent to SMS messages from Enveda Therapeutics. The website privacy policy states under Your Rights that "we do not collect personal data", although those forms collect it.

No retention period or handling standard for screening answers is published.

Enveda sponsors trials of three medicines and recruits for its atopic dermatitis and asthma studies through its own pages, where visitors answer eligibility questions and leave contact details. The trial pages tell participants that personal information will be kept confidential and de identified whenever possible, and that the study doctor, the sponsor, the FDA and the institutional review board may inspect records that identify a participant by name. Its website privacy policy, effective November 6, 2024, covers the website, its contact form, cookies and an SMS program whose consent data may be shared when needed to advance participation in a clinical trial.

No statement on business associate status, or on how the privacy rule applies to the company, is published.

Enveda's systems hold its spectral library and pipeline chemistry, and as a trial sponsor it can inspect study records that identify participants by name and receives the screening answers and contact details that the public enters on its recruitment pages. Its website privacy policy commits it to appropriate security measures that safeguard personal information from unauthorized access, disclosure or misuse, and states that it cannot guarantee absolute security.

No security certification, trust center or independent assessment is named.

Investigational new drug applications have been cleared for three Enveda medicines: ENV-294, whose advance into clinical trials was announced on November 13, 2024, ENV-308, cleared with a first patient dosed in December 2025, and ENV-6946, cleared with Phase 1 initiated in December 2025. The Phase 2a atopic dermatitis study is registered as NCT07298395, and all three molecules remain investigational agents not approved for use by any regulatory authority.

PRISM and the rest of the discovery platform are used inside the company rather than sold, and carry no medical device claim.

PRISM pretraining was tested in a controlled comparison that nine named authors published on May 28, 2024: the same model architecture trained on the same labeled data with and without PRISM pretraining. It reports a relative increase in R squared of 7 to 16 percent across twelve chemical properties and a 23.3 percent relative improvement in retrieving the closest molecule from a spectral reference library, against a baseline trained on roughly 2 million annotated spectra from about 60,000 reference molecules, and performance kept rising with training set size rather than leveling off.

For the natural material behind the library, Enveda states that it follows access and benefit sharing provisions in line with the Convention on Biological Diversity and works on biodiversity conservation and restoration with global partners.

No result is broken down by plant taxon, geography, instrument or chemical class.

Laboratory work and the drug development process stand between a model prediction and a patient. Molecules that PRISM proposes are tested in Enveda's automated laboratory and refined by its chemists before any becomes a development candidate, and a candidate cannot reach a human subject without an investigational new drug clearance and institutional review board approval. The recruitment page for the Phase 2a atopic dermatitis trial, NCT07298395, tells participants that an independent review board reviews the study and that doctors will monitor their health closely.

No error rate for PRISM's predictions is published.

The platform works on mass spectrometry data inside Enveda's own research, reading public metabolomics data from GNPS, MetaboLights and the Metabolomics Workbench alongside spectra from its own laboratory. Clinical work runs through research sites such as JDR Dermatology Research in Las Vegas and Arlington Research Center in Arlington, Texas, and the platform serves Enveda's own programs rather than providers or health systems.

No connection to health record systems, laboratory information systems or electronic notebooks is described.

Microsoft provides the compute behind PRISM: the model was built with Azure, and Enveda uses Azure to scale its generative AI models. The platform serves Enveda's own research rather than customers, and the company works from Boulder, Colorado and from Genome Valley, Hyderabad, where Enveda Therapeutics India Pvt Ltd is based.

No Azure region or data residency position is published.

Venture financing pays for Enveda's pipeline, and each round is announced with its size and investors: $51 million in Series A in June 2021 led by Lux Capital, $68 million in combined Series B equity and debt in December 2022, an oversubscribed Series B1 bringing that round to $119 million in April 2023, $55 million in June 2024, a $130 million Series C in November 2024 led by Kinnevik and FPV and extended to $150 million in February 2025 when Sanofi joined, a $150 million Series D led by Premji Invest in September 2025, and a $311 million Series E led by Catalio Capital Management on September 23, 2026. Total capital raised since inception is more than $845 million. The Series E is to carry ENV-294 and ENV-308 into later stage trials and additional indications, take ENV-6946 through mid phase trials, bring more medicines into the clinic and scale PRISM and the automated laboratory that feeds it.

No partnering, licensing or collaboration terms are published for any program.

Three oral medicines in human studies cover four named indications: moderate to severe atopic dermatitis and asthma for ENV-294, maintenance of weight loss and metabolic health for ENV-308, and inflammatory bowel disease, including ulcerative colitis and Crohn's disease, for ENV-6946. Each has a named mechanism: a non kinase inflammatory pathway for ENV-294, which Enveda describes as a molecular glue, a mimic of an exercise hormone acting on satiety and reward pathways for ENV-308, and a gut preferred inhibitor of TNF alpha, IL-23 and TL1A signaling for ENV-6946.

Beyond those programs, Enveda works with the Bill & Melinda Gates Foundation to search natural chemistry for medicines for tuberculosis, malaria and other global health priorities, has run a survey on dog health, and names diagnostics, agriculture, nutrition and manufacturing as later uses for the platform.

Indications for the rest of the seventeen development candidates are not published.