Deep Genomics

There is nothing else here. The company presents no pipeline, no clinical asset and no service other than what its models produce, and it now describes itself around the foundation model platform rather than around a therapeutic ambition. Three distinct models are named and each does a different job, from variant effect prediction to guide design for editing.

The laboratories exist to generate training and validation data for those models rather than to run a discovery operation the models assist. This is the purest case of the grade in the lane, and it is pure partly because everything that was not the platform has fallen away.

The validation architecture is described concretely and backed by physical capacity. The company operates lab in the loop workflows across two facilities totalling more than 10,000 square feet in Toronto and Cambridge, generates datasets fit to the purpose of training or fine tuning rather than reusing whatever exists, and validates predictions using orthogonal assays including functional assays before feeding results back into training.

Orthogonal validation is the substantive commitment there: checking a prediction by a method that does not share the original's assumptions is the honest form of the test, and the model papers carry it through to published task suites. What is absent is the same thing missing across this lane, a statement of human decision authority, and unlike two neighbours here the company publishes no round by round hit rates from inside the loop.

The most complete technical picture of a deployed system graded in this lane, and it earns the grade on one thing nobody else offers: a version history. The flagship model is disclosed at close to two billion tunable parameters trained on over a trillion signals, dated to initial development in 2021, with its evolution stated specifically, including a shift in prediction resolution from 128 base pairs down to a single base pair, the addition of disease context training data, and architectural changes for predictive power and efficiency.

The second model's task suite and comparative claims are described, and the third's training regime is given down to the count of endogenous editing sites observed and the fine tuning on a custom synthetic screening dataset. Each has a public preprint. Almost every vendor in this index describes what a model does; this one describes what changed and when. Withheld: the weights, the identity of the training datasets, and peer reviewed publication of the flagship work, which has sat as a preprint since September 2023.

Among the best in this lane, on specifics rather than on policy. A third party dependency is named outright, with an experiment tracking platform integrated into the flagship model's workflows, and naming an actual vendor in the machine learning stack is something almost no company in this index does.

Training inputs are characterised by scale and type rather than waved at, at over a trillion signals from high throughput sequencing for one model and sixteen million observed editing sites plus a custom synthetic screening dataset for another, and the models are described as trained from scratch or fine tuned rather than adapted from an unnamed base.

What remains absent is the provenance that matters most: no dataset or consortium is named as a source, no compute or hosting provider is disclosed, no open source component inventory exists, and no data use terms are stated for material that originated with human participants.

The grade rests almost entirely on one half of the axis. The scientific half is the strongest in this lane: seventeen listed publications including Science, Nature Biotechnology twice, Nature Medicine and Molecular Systems Biology, with model specific work reporting checkable results, including a study across fourteen genes in which the flagship model designed effective tissue specific steric blocking oligonucleotides for targets including Wilson disease and spinal muscular atrophy.

The operational half is close to empty. No pharmaceutical partner is named anywhere, no collaboration or programme outcome is disclosed, and no molecule is in the clinic. Two qualifications belong with the science. The landmark peer reviewed papers are largely the founder's academic lineage from 2015 to 2018, while every paper describing the current commercial models sits on a preprint server, the flagship one unpublished in a peer reviewed venue for nearly three years as of 29 August 2026. And the most quoted external endorsement came from a member of the company's own scientific advisory board.

This vendor carries a governance question no other company in this lane faces, and addresses it nowhere. Its flagship model was trained on over a trillion signals from high throughput sequencing, including paired genotype and RNA expression data from many individuals, and a second model on sixteen million endogenous editing sites.

Human genomic and transcriptomic data of that kind normally arrives through research consortia carrying consent terms and data use agreements that restrict downstream commercial application. Nothing published names the sources, states the consent basis, or explains how any such obligations travel into a model now offered commercially to pharmaceutical partners.

A model is not a dataset, and whether restrictions follow the weights is an unsettled question that this company's business model puts squarely in play. Recorded as a located absence: nothing indicates a problem, and nothing addresses the question either. The partner separation issue common to this lane also applies, though with no partners named it cannot be assessed.

Graded neutrally, and the neutrality is less comfortable here than elsewhere in this lane. The company is Canadian, its data handling falls under Canadian and European regimes rather than the United States framework this axis was written against, no covered entity relationship exists and no business associate agreement would apply.

But unlike the chemistry and antibody vendors graded alongside it, this platform is not built on molecules alone: the flagship model learns from paired genotype and RNA expression data across many individuals, which is human derived material. The obligation this axis measures does not attach, and the underlying concern it exists to capture does. That concern is graded on the stewardship axis rather than dismissed here.

Two dedicated passes on 29 August 2026 across a targeted search and a direct review of all seven navigation sections and the site footer found no security page, no trust centre, no certification and no attestation. The footer offers a privacy policy and a cookie preferences control and nothing else.

The absence lands harder here than at the chemistry vendors in this lane: this company's training corpus includes human genomic and transcriptomic data, its models are the asset, and it publishes nothing about how either the data or the weights are protected.

Nothing to clear. No candidate is in the clinic, no pipeline is currently presented, no investigational application is disclosed, and two dedicated passes on 29 August 2026 located no enforcement action or adverse standing. One item deserves recording on the credit side even though it does not move the grade: the company contributed to updated industry recommendations on assessing hybridisation dependent off target risk for therapeutic oligonucleotides, published in a peer reviewed journal. Helping write the safety guidance for its own field is a different and better signal than complying with it, and almost nothing else in this lane does anything comparable.

No governance framework, model card or responsible use policy appears across the seven sections of the site as of 29 August 2026. The substitution is more substantial than at most neighbours. Evaluation methodology is documented in the published work, including comparison against other architectures on orthogonal task suites and out of distribution prediction from massively parallel reporter assays, which is a harder test than in distribution accuracy.

The platform description names model evaluation frameworks as a component rather than an afterthought, and the off target risk recommendations paper is governance work in substance if not in form. What none of that supplies is the thing a partner would need: any statement of when the company judges a model unfit to run a programme, or what it does when a prediction and an assay disagree.

No terms of service exist on any public surface as of 29 August 2026. The only legal document is a website privacy policy, joined by a cookie preferences control. There is no warranty position, no limitation of liability, no indemnity, no service level and no allocation of rights over model outputs or over anything derived from them.

For a company whose entire commercial proposition is now supplying model predictions into partner drug programmes, the question of who carries the consequence when a prediction is wrong is the central one, and nothing public addresses it.

No electronic health record surface exists and none would be appropriate, so this is graded neutrally under the convention for this lane. Recorded in its place: the platform is explicitly designed to be consumed by other systems, generating sequence embeddings intended as inputs to downstream models used in target biology and molecular design, and the workflows support fine tuning on new datasets, which is a genuine interoperability posture rather than a closed pipeline.

The tooling around it is named, unusually. What does not exist is any product surface, application programming interface or connector a partner could integrate independently, because access appears to run through collaboration rather than through software.

No residency, tenancy, hosting or retention statement exists for partner data, and no software is deployed to anyone. Graded neutrally under the convention for this lane, now the seventh consecutive record where that convention rather than the evidence sets this grade. Two facts sit here for a later reader.

The company names a third party experiment tracking platform integrated into its model workflows, which is a processor a partner would want listed and is more than any neighbour discloses. And this is a Canadian headquartered company operating a United States facility and training on human derived genomic data, which raises a cross border data question with real regulatory content behind it. Nothing published touches it.

The lowest grade on this axis in the lane and the only D, awarded on absence rather than on evasion. The company states that its platform advances drug discovery problems both internally and in support of major pharmaceutical partners, and it names none of them. There is no collaboration announcement, no deal value, no upfront, no milestone structure, no licensing arrangement and no customer of any kind on the site or in the news archive.

Neither is there a funding figure, an investor list or a pipeline. Every other vendor graded in this cluster names at least three partners and most disclose a deal value. Compounding it, the strategic direction announced in September 2023, to become a forward integrated biopharmaceutical company advancing multiple therapeutics, is no longer reflected anywhere, and two dedicated passes on 29 August 2026 located no announcement of the change. A reader cannot establish what this company sells, to whom, or on what terms.

Broader than the antibody vendors in this lane, within a single modality. The three model families reach different mechanisms rather than variations of one: sequence to expression regulation and variant effect, microRNA binding and messenger RNA degradation, and guide design for ADAR mediated editing. Applications named span target biology discovery, molecular design, on and off target liability prediction for genetic medicines, and embeddings feeding downstream models.

Historical work covers splicing, polyadenylation and steric blocking oligonucleotides. What limits the grade is the far end: no clinical stage, no named indication currently in play, and everything sits inside RNA, so a partner working on any other modality has nothing to buy here.