Atlas Discovery
AI-native pharma company building foundation models and agents that predict patient drug response and de-risk clinical trials
The dapagliflozin rank-2 versus TxGNN's 4,388 is a real headline, and RepurposingBench — where Atlas publishes prospective scores under 9/100 — tells you the team is honest about what these models can't yet do. scMDM's 83.3% label-free responder accuracy in melanoma and the AUROC 0.858 trial-success model are close behind. If you are a pharma or translational team holding response-labeled clinical data, that conversation is worth having now; if you need a validated diagnostic output or an off-the-shelf prediction endpoint, this is one to watch, not buy.
Verified 15d ago · liveness 61/100 · cite: rightaichoice.com/tools/atlas-discovery
- Pharmaceutical R&D teams running drug repurposing screens against human response models
- Clinical trial designers modeling patient selection and enrollment power before Phase II/III
- Translational bioinformaticians working with single-cell and multi-omics patient data
- Rare-disease foundations seeking repurposing candidates, as with Alliance to Cure CCM
- Chemists focused on small-molecule synthesis and lead optimization
- Groups without response-labeled clinical or trial data to bring to a collaboration
- Anyone needing a validated diagnostic or clinical decision output they can act on directly
We scan live Reddit threads, YouTube comments, X posts, G2 reviews and other communities — and hand you an honest verdict in under a minute.
- Honest verdict, not marketing
- Real pros & cons from real users
- Attributed quotes with receipts
3 free scans · no card needed
Skip Atlas Discovery if you need a validated diagnostic or clinical decision output you can act on directly, or if you have no response-labeled clinical or trial data to bring to a research collaboration.
Benchmark scores are the honest ceiling, not the floor: RepurposingBench prospective scores sit under 9/100, so a repurposing shortlist still needs wet-lab validation budget behind it.
The shape of the engagement — research collaboration tied to your own response-labeled clinical data — puts it in the same budget conversation as custom computational-biology and translational-research partnerships rather than a seat-based SaaS subscription. Evaluate it against the cost of the trial enrollment you are trying to avoid: a 458-patient reduction in UNIFI enrollment is the
In short
Atlas Discovery — AI-native pharma company building foundation models and agents that predict patient drug response and de-risk clinical trials. Best for Pharmaceutical R&D teams running drug repurposing screens against human response models, Clinical trial designers modeling patient selection and enrollment power before Phase II/III, Translational bioinformaticians working with single-cell and multi-omics patient data. Contact Sales pricing.
What's new in Atlas Discovery
Checked 4 days agoAcross the latest 6 updates: 1 community discussion and 5 news mentions.
Pushing the Economic Frontier of Medicine
Atlas Discovery argues AI shifts knowledge work to inference cost, enabling agent-driven firms to target long-tail diseases where biology is tractable but market size never justified development.
RepurposingBench: A Benchmark for Drug Repurposing
RepurposingBench separates reasoning from recall by holding out drug-disease pairs absent from pretraining. Weighted recall lands at 35-41% on rediscovery; all models score under 9/100 on the prospective set.
Agents for Drug Repurposing
Specialized agent harness builds representations per hypothesis instead of frozen knowledge graphs. Backtested on a frozen corpus, it placed all three SGLT2 inhibitors in the top 3 for heart failure.
ClinicBench: Evaluating Frontier Models on Clinical Tasks
ClinicBench draws on 500,000+ patient records to test 30-day care forecasting. Best of 9 frontier models scores 51.5/100, and all favor plausible continuations over actual care.
Clinical Trial Prediction with Agents
Agent harness assembles scattered evidence and predicts clinical success probabilities, beating leading academic baselines on AUROC and AUPRC at every phase transition; 0.858 AUROC for Phase III to approval.
Atlas Discovery Partners with Alliance to Cure Cavernous Malformation
A repurposing candidate found by Atlas research agents enters preclinical testing with Alliance to Cure CCM, a disease affecting 18-24 million people with no approved cure.
What people actually say about Atlas Discovery — is it worth it?
We scanned public community sources for Atlas Discovery on Aug 16, 2026 and could not establish that the discussion we found is about this tool rather than something else sharing its name. Our own analysis of that scan says the posts were off-subject. Rather than publish a sentiment score built on the wrong subject, we publish nothing here and re-run the scan.
Viability Score
How well maintained and how widely used is Atlas Discovery? Built from what the vendor actually publishes (docs, changelog, tutorials, integrations, pricing), whether the site is live, and how much real users discuss it. How we calculate this
Last calculated: October 2026
How we score →Key Features
- Foundation models predicting individual patient drug response from pre-clinical and clinical data
- Ustekinumab response in ulcerative colitis predicted from baseline biopsy at AUROC 0.76
- ExpressionVAE discrete-token encoding for single-cell data, 3–20x better than continuous baselines
- scMDM masked discrete diffusion predicts treatment responders without response labels, 83.3% melanoma accuracy
- Agent harness for drug repurposing that builds hypothesis-specific representations
- Recovers dapagliflozin at rank 2 for heart failure vs TxGNN's 4,388th of 7,957
- Agent-based clinical trial success prediction at AUROC 0.858, Phase III to approval
- Virtual patient cohorts and power analysis for in silico trial design
- Estimated UNIFI enrollment reduction of 458 patients via response prediction
- ClinicBench benchmark forecasting next 30 days of care from 500k+ patient records
- RepurposingBench with direct-probing held-out pairs and post-cutoff disclosures
- Multi-omics integration linking patient biology to drug outcomes
- Published research at ICLR '26, CSHL '26, and ICML '26
- Partnered drug repurposing candidate in preclinical testing with Alliance to Cure CCM
About Atlas Discovery
Atlas Discovery is an AI-native pharma company building autonomous agents and foundation models that predict how individual patients will respond to drugs, then uses those predictions to repurpose existing medicines and design more efficient clinical trials. The company's framing of the problem is blunt: nine of ten drugs fail in trials because animal and cell models don't reliably forecast human response, and the data that would do better sits in disconnected silos that never tie a patient's biology to a drug outcome. The technical core is ExpressionVAE, a discrete-token encoding method for single-cell data that Atlas reports beats continuous latent-variable baselines by 3–20x on distributional metrics. In published clinical validation, its foundation model predicted ustekinumab response in ulcerative colitis from baseline biopsies at an AUROC of 0.76 — a result Atlas says could have cut UNIFI trial enrollment by 458 patients. A 2026 release adds scMDM, a masked discrete diffusion model that separates responders from non-responders without any response labels, hitting 83.3% accuracy in a melanoma trial. Newer work pushes into agentic territory: a specialized agent harness for drug repurposing builds hypothesis-specific representations and recovers dapagliflozin at rank 2 for heart failure, against TxGNN's 4,388th out of 7,957 compounds. The same agent approach predicts clinical trial success from scattered evidence at AUROC 0.858 for Phase III to approval. Atlas also publishes benchmarks rather than only claims — ClinicBench (500k+ patient records, best frontier score 51.5/100) and RepurposingBench, which grades weighted recall at 35–41% and prospective scores under 9/100. In August 2026 a drug repurposing candidate entered preclinical testing with Alliance to Cure Cavernous Malformation. Pharma R&D and translational teams holding their own response-labeled clinical datasets are the natural fit; access is arranged by contacting the Atlas team directly.
Behind the Verdict
Atlas Discovery is not a dashboard you log into; it is a research organization that sells collaboration. The pitch is that most drug failures trace back to models that don't predict human response, and that the fix is better representation learning on the patient data you already have. Two things give that pitch credibility. First, the methods are specific and named: ExpressionVAE's discrete-token encoding for single-cell data, reported 3–20x better than continuous latent-variable baselines on distributional metrics, and scMDM, a masked discrete diffusion model that called responders without response labels at 83.3% accuracy in a melanoma trial. Second, the results are attached to numbers rather than adjectives — ustekinumab response from baseline biopsy at AUROC 0.76, with an estimated 458-patient reduction in UNIFI enrollment, and dapagliflozin recovered at rank 2 versus TxGNN's 4,388th of 7,957. The agent harness work is the more interesting bet: a system that builds hypothesis-specific representations per target, and a second agent that turns scattered evidence into clinical-success probabilities at AUROC 0.858 for Phase III to approval. Where Atlas stands apart from most AI-pharma marketing is that it publishes the benchmarks that make it look bad. ClinicBench asks models to forecast the next 30 days of care from 500k+ patient records; the best frontier score is 51.5/100, and the models turn out better at plausible care than actual care. RepurposingBench grades weighted recall at 35–41% and prospective scores under 9/100. Those numbers are the honest floor of what this class of model can do today, and a team that publishes them is a team you can calibrate against. The constraints are equally real. The only disclosed repurposing candidate, with Alliance to Cure Cavernous Malformation, is an FDA-approved drug with 10+ years of safety data that just entered preclinical testing in August 2026; no readout exists yet. If your work is small-molecule synthesis or lead optimization, this isn't your tool. If you need a validated diagnostic or a clinical decision you can act on tomorrow, this isn't that either. The right way to read Atlas is as a research partner with unusually transparent benchmarks and a small but real set of validated wins — worth a scoping call if you hold the data, and worth tracking if you don't.
Researching Atlas Discovery? Get your full AI stack in 60 seconds.
Free, no signup — tell us your goal and get tools matched to your budget & existing stack.
Real-world workflow fit
Concrete scenarios for the personas Atlas Discovery actually fits — and what changes day-one when you adopt it.
You have baseline biopsy single-cell and multi-omics data plus response labels from a completed trial. You hand Atlas a cohort to test whether ExpressionVAE and the response-prediction foundation model reproduce the treatment effect signal in your indication, mirroring the ustekinumab AUROC 0.76 result.
Outcome: You get a first read on whether a patient-selection biomarker is learnable from your baseline data, which decides whether the next trial design can drop enrollment the way Atlas estimated a 458-patient reduction for UNIFI.
Before locking a sample size, you use Atlas's virtual patient cohorts and power analysis to model how much enrollment drops if responders are enriched at baseline.
Outcome: You arrive at the protocol review with a defensible, model-backed enrollment number rather than a rule-of-thumb inflation, and you can see where the model is uncertain.
You bring a disease area with an approved drug and 10+ years of safety data but no clear indication match, and you ask the agent harness to rank repurposing candidates — the same setup that surfaced dapagliflozin at rank 2 for heart failure.
Outcome: You get a shortlist with hypothesis-specific representations behind each candidate and a concrete path to preclinical testing, which is the stage the Alliance to Cure CCM candidate is at now.
Use Cases
- Predict patient drug response from baseline biopsies to inform trial enrollment.
- Rescue failed drugs by re-evaluating their clinical data with foundation models.
- Design trials with fewer patients using power analyses from virtual cohorts.
- Integrate pre-clinical and clinical data silos into unified patient biology models.
- Identify new indications for existing drugs using the agent harness for drug repurposing.
Limitations
- Model performance is reported on research benchmarks and specific case studies rather than a generally available product.
- RepurposingBench weighted recall is 35–41%, every model scores under 9/100 on the prospective set, and the best of 9 frontier models scores 51.5/100 on ClinicBench, where models proved better at plausible care than actual care. scMDM's 83.3% responder accuracy comes from a single melanoma trial.
- The disclosed drug repurposing candidate, with Alliance to Cure Cavernous Malformation, is an FDA-approved drug with 10+ years of safety data that entered preclinical testing in August 2026 — no clinical readout exists.
- Natural fit is teams with their own response-labeled clinical or trial data.
as of 2026-09-23
Verification history
We have re-verified Atlas Discovery 8 times since . Each pass re-reads the vendor's own pages and re-checks every listed field against that evidence; passes where nothing had changed are marked as such.
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
- — re-verified summary, description, our verdict, our analysis, pricing model, pricing tiers, features, integrations, who it suits, who should skip it
Showing the 6 most recent of 8 verification passes.
Free to cite with attribution — this page re-verifies continuously.
Where the pricing makes sense
The company stage and team size where Atlas Discovery's pricing actually pencils out — and where peers do it cheaper.
The shape of the engagement — research collaboration tied to your own response-labeled clinical data — puts it in the same budget conversation as custom computational-biology and translational-research partnerships rather than a seat-based SaaS subscription. Evaluate it against the cost of the trial enrollment you are trying to avoid: a 458-patient reduction in UNIFI enrollment is the
Setup time & first value
How long it actually takes to get something useful out of Atlas Discovery — broken out by persona, not the marketing-page minute.
For pharma and translational teams, first value comes from a scoping conversation rather than a signup, then data-sharing and cohort definition before any model runs — expect weeks, not hours, given the clinical data involved. A translational bioinformatician already holding clean, labeled cohorts moves fastest, since the model input is baseline biopsy and multi-omics data you likely already
Switching to or from Atlas Discovery
How to bring data in from common predecessors and how to get it back out — written for the switcher, not the buyer.
- →From TxGNN-style repurposing baselines: re-run your candidate screen through the agent harness, which recovered dapagliflozin at rank 2 for heart failure against TxGNN's 4,388th of 7,957.
- →From continuous latent-variable single-cell encoders: re-encode your datasets with ExpressionVAE's discrete-token method, reported 3–20x better on distributional metrics.
- →From academic trial-success models: feed the same evidence set to the agent-based predictor that reached AUROC 0.858 for Phase III to approval.
- ↗To an in-house foundation model team: take the ExpressionVAE and scMDM method descriptions and rebuild against your own compute, accepting the benchmark gaps RepurposingBench documents.
- ↗To a wet-lab CRO: once the agent harness shortlists candidates, the work shifts to synthesis and assay validation, which is outside Atlas's scope.
Resources & Guides
Tutorials & Learning
YouTube returned 5 videos for “Atlas Discovery”, and we withheld 5: 5 did not mention Atlas Discovery. We are showing none, because we could not prove any of them are about Atlas Discovery.
Official links
Tools that pair well with Atlas Discovery
Common stack mates teams adopt alongside Atlas Discovery, with the specific reason each pairing earns its keep.
Verge Genomics
All-in-human AI foundation models that predict brain activity and therapy response for CNS drug programs
Recursion
Clinical-stage AI drug discovery company running a wet-and-dry lab-in-the-loop engine with 2M weekly experiments and a 50+ PB phenomics dataset
Insilico Medicine
Generative AI drug discovery suite covering target ID, molecule design, biologics engineering and clinical trial prediction.
Featured Head-to-Head Comparisons
Atlas Discovery vs Codametrix
Choose CodaMetrix if you need enterprise medical coding automation with proven ROI and deep EHR integration for large health systems. Choose Atlas Discovery if you are in pharma R&D aiming to de-risk clinical trials with foundation models that predict patient drug response. They address entirely different domains—no direct competition.
Atlas Discovery vs Isomorphic Labs
Isomorphic Labs is the pick for large pharma seeking end-to-end AI drug discovery with AlphaFold prestige and massive funding. Atlas Discovery is better for organizations focused on de-risking clinical trials via patient response predictions. Choose based on your bottleneck: molecule design or clinical success prediction.
Atlas Discovery vs Rapidsos
RapidSOS and Atlas Discovery operate in completely different domains. Choose RapidSOS if you need AI-powered emergency response infrastructure for 911 centers or enterprise safety. Choose Atlas Discovery if you are in pharma R&D and want to de-risk clinical trials with patient response predictions. They are not competitors.
Alternatives to Atlas Discovery
View all
Verge Genomics
All-in-human AI foundation models that predict brain activity and therapy response for CNS drug programs
Recursion
Clinical-stage AI drug discovery company running a wet-and-dry lab-in-the-loop engine with 2M weekly experiments and a 50+ PB phenomics dataset
Insilico Medicine
Generative AI drug discovery suite covering target ID, molecule design, biologics engineering and clinical trial prediction.
Frequently Asked Questions
Categories
Topics
Used Atlas Discovery? Help shape our editorial sentiment research.