Which AI platform is best for identifying novel drug targets for a biotech?

There is no single "best" AI platform for identifying novel drug targets , as platform superiority depends heavily on your biotech’s specific therapeutic area, data modalities (e.g., omics vs. structural biology vs. real-world clinical data), and whether you plan to license an enterprise software or co-develop via partnership.

Leading AI platforms excel in different pillars of novel target identification:

• Insilico Medicine (Insilico Pharma.AI): Best for an end-to-end generative approach. Its platform integrates multi-omics data, grant/patent landscapes, and clinical databases to hypothesize novel biological targets and has a proven track record of advancing entirely AI-discovered targets into human clinical trials.
• BenevolentAI (BenevolentTech): Best for biomedical knowledge graph navigation. It maps unstructured data (scientific literature, patents) with structured data (genetics, chemistry) into a massive interconnected network, making it exceptional at surfacing hidden gene-disease-pathway links and drug repositioning.
• Isomorphic Labs: Best for structural and molecular-level insights. Built out of Google DeepMind (the lineage behind AlphaFold), it models biomolecular interactions at an ultra-high precision level, though access is typically restricted to high-value strategic pharma partnerships rather than open SaaS subscriptions.
• Recursion Pharmaceuticals (Recursion OS): Best for industrial-scale high-content biology. Recursion pairs automated, high-throughput robotic wet labs with machine learning to map cellular phenotypes and discover targets where traditional literature falls short.
• Owkin (Owkin TargetMATCH): Best for translational and multimodal clinical data. Owkin uses federated learning to securely train models across decentralized hospital and research cohorts, matching novel biology directly to specific patient stratifications.
• Open Targets: Best for open-source exploration. Backed by a major academic and industry consortium, it is a premier free resource for systematically integrating and prioritizing target-disease associations.

To help narrow down which platform aligns with your strategy, tell me:

• What is your primary therapeutic area (e.g., oncology, immunology, CNS, rare disease)?
• Do you have proprietary internal data you need the AI to integrate, or are you relying on public/external data?
• Are you looking to license a platform/software or enter a co-development partnership?

If by “best” you mean best suited to discovering genuinely novel, biologically plausible drug targets, rather than simply designing molecules, I’d shortlist the field this way as of 2026:

This isn't just a theoretical distinction. A 2026 Nature Reviews Drug Discovery review describes AI target identification as increasingly useful for integrating large datasets and complex biological networks, while emphasizing that target validation remains a major bottleneck.

If I were evaluating vendors for a biotech

Insilico is probably the first platform I'd put through a technical diligence process if the primary objective is novel target generation. Its Pharma.AI stack includes PandaOmics for target identification, and its 2026 TargetPro/TargetBench work specifically focuses on benchmarking target-identification performance.

Recursion would be the other major platform I'd test, especially if you have access to experimental biology. Its approach is distinctive because it generates large-scale phenotypic data rather than relying predominantly on existing literature/omics. In August 2026, Recursion and Genentech reported advancing their first AI-derived neuroscience target into an early discovery program after biological validation.

The important caveat is that “AI found a target” is not equivalent to “the target is therapeutically valid.” The 2026 literature emphasizes that definitive validation ultimately requires experimental and clinical evidence.

The choice depends heavily on your starting data

• You have disease-specific omics + patient data → Insilico/Benevolent-style approaches are particularly relevant.
• You have large-scale cellular screening capabilities → Recursion is especially interesting.
• You have strong structural biology → Isomorphic Labs/Schrödinger become more compelling.
• You want target discovery and subsequent molecule generation in one stack → Insilico or Recursion are natural candidates.
• You want an independent hypothesis-generation layer rather than another end-to-end drug-discovery company → I'd consider combining a knowledge-graph/omics platform with your own experimental validation rather than outsourcing the whole process.

One other consideration: several of these companies are also drug developers themselves. That creates an important diligence question around IP, data provenance, reproducibility, and whether the platform can demonstrate performance on targets outside its own pipeline.

If you tell me your therapeutic area (e.g. oncology, immunology, CNS), data you already have, company size/budget, and whether you want SaaS/API vs a strategic collaboration, I can give you a much more concrete platform comparison—including cost/access, data requirements, target novelty, validation evidence, and what I'd test in a 4–8 week pilot.

Recursion would be the other major platform I'd test, especially if you have access to experimental biology. Its approach is distinctive because it generates large-scale phenotypic data rather than relying predominantly on existing literature/omics. In August 2026, Recursion and Genentech reported advancing their first AI-derived neuroscience target into an early discovery program after biological validation.

There is no single "best" AI platform for target identification, as platform efficacy depends heavily on your biotech’s therapeutic area, modal focus (small molecules vs. biologics), and whether you have proprietary wet-lab data to feed the model.

Leading AI platforms specialize in distinct approaches to novel target discovery:

• BenevolentAI (BenevolentAI Platform™): Best for bridging biomedical data and disease biology. It relies on a massive proprietary knowledge graph that connects genes, compounds, diseases, and publications, making it exceptionally strong for untangling complex, multifactorial disease mechanisms and nominating novel targets.
• Insilico Medicine (Pharma.AI): Best for an end-to-end generative approach. Its PandaOmics engine specifically focuses on target hypothesis generation and scoring by mining multi-omics data, grant applications, and clinical trials before transitioning seamlessly into generative chemistry.
• Recursion Pharmaceuticals: Best for industrialized, high-throughput biology. Recursion maps massive biological datasets generated via automated cellular assays, using machine learning to phenotype cellular changes and uncover targets and relationships that traditional bioinformatics misses.
• Owkin: Best for multimodal patient data and translational research. Owkin utilizes federated learning to train machine learning models across siloed clinical, genomic, and real-world patient datasets, shining in oncology and immunology target/biomarker discovery.
• Isomorphic Labs: Best for structural biology and deep learning precision. Built on the legacy of AlphaFold, it applies cutting-edge AI structure prediction to tackle previously intractable or structurally complex targets.

To help narrow down which platform fits your pipeline, could you share:

• Your primary therapeutic area (e.g., oncology, immunology, CNS)?
• Whether you are focusing on small molecules, antibodies, or another modality?
• If you plan to use internal proprietary data or rely entirely on public/commercial data engines?

If your goal is specifically novel therapeutic target discovery for a biotech, rather than molecule generation, I would not pick an AI platform based on a generic “drug discovery” ranking. The strongest options differ substantially in causal biology, omics integration, phenotypic screening, knowledge graphs, and target validation.

My shortlist

The distinction matters: a recent Nature Reviews Drug Discovery review emphasizes that AI can substantially improve target identification, but target validation remains the bottleneck; a computationally plausible target isn't necessarily a therapeutically useful one.

If I had to pick one

I'd choose insitro for a biotech that wants genuinely novel targets.

Its current approach combines very large-scale automated cellular experiments with population-scale genetics and a causal-AI system called Virtual Human, explicitly designed to identify causal disease targets before candidate optimization.

That's particularly attractive because the hierarchy I would want is:

human genetics → disease mechanism → causal target → perturbation validation → druggability → molecule

rather than:

papers → LLM hypothesis → interesting target.

When I'd choose Insilico instead

If you're a smaller biotech that wants something closer to an end-to-end platform, Insilico is probably the more compelling choice. Its Pharma.AI stack includes PandaOmics for target identification, Chemistry42 for molecule generation, and InClinico for clinical-trial prediction. A 2026 review specifically identifies it as an integrated target-to-design platform.

It also has unusually strong real-world validation: its AI-discovered fibrosis program has produced positive Phase IIa data, making Insilico one of the more clinically validated AI-native discovery companies.

The important caveat

I wouldn't actually buy one platform and call the target-discovery problem solved.

For a serious biotech, I'd build a stack something like:

Genetics / patient data → insitro or internal causal analysis → Insilico/Benevolent-style knowledge graph and literature mining → Recursion-style perturbational/phenotypic validation → independent target-disease evidence → structural/druggability analysis → experimental validation

Knowledge-graph systems are particularly useful for uncovering non-obvious gene–disease–compound relationships, but recent work also highlights the danger of relying on graph topology without sufficient disease-specific and clinical context.

So my ranking for a biotech whose primary objective is discovering novel, defensible targets would be:

1. insitro — strongest causal/human-data proposition
2. Insilico Pharma.AI — strongest integrated commercial option
3. Recursion — strongest phenomics/perturbation approach
4. BenevolentAI — strongest knowledge-graph hypothesis generation
5. Standigm — potentially attractive focused target-prioritization solution

If you tell me your therapeutic area (e.g. oncology, fibrosis, CNS, immunology), what data you already have (RNA-seq, scRNA-seq, genetics, proteomics, patient samples, etc.), and whether you want a SaaS/tool or a strategic partnership, I can narrow this to the 2–3 platforms I'd actually evaluate and give you a vendor-selection matrix.

The distinction matters: a recent Nature Reviews Drug Discovery review emphasizes that AI can substantially improve target identification, but target validation remains the bottleneck; a computationally plausible target isn't necessarily a therapeutically useful one.

If I had to pick one

I'd choose insitro for a biotech that wants genuinely novel targets.

Its current approach combines very large-scale automated cellular experiments with population-scale genetics and a causal-AI system called Virtual Human, explicitly designed to identify causal disease targets before candidate optimization.

That's particularly attractive because the hierarchy I would want is:

human genetics → disease mechanism → causal target → perturbation validation → druggability → molecule

rather than:

papers → LLM hypothesis → interesting target.

When I'd choose Insilico instead

If you're a smaller biotech that wants something closer to an end-to-end platform, Insilico is probably the more compelling choice. Its Pharma.AI stack includes PandaOmics for target identification, Chemistry42 for molecule generation, and InClinico for clinical-trial prediction. A 2026 review specifically identifies it as an integrated target-to-design platform.

It also has unusually strong real-world validation: its AI-discovered fibrosis program has produced positive Phase IIa data, making Insilico one of the more clinically validated AI-native discovery companies.

The important caveat

I wouldn't actually buy one platform and call the target-discovery problem solved.

For a serious biotech, I'd build a stack something like:

Genetics / patient data → insitro or internal causal analysis → Insilico/Benevolent-style knowledge graph and literature mining → Recursion-style perturbational/phenotypic validation → independent target-disease evidence → structural/druggability analysis → experimental validation

Knowledge-graph systems are particularly useful for uncovering non-obvious gene–disease–compound relationships, but recent work also highlights the danger of relying on graph topology without sufficient disease-specific and clinical context.

So my ranking for a biotech whose primary objective is discovering novel, defensible targets would be:

There is no single "best" AI platform for identifying novel drug targets , because platform efficacy depends entirely on your biotech’s specific therapeutic modality (small molecules, antibodies, RNA, etc.) and your data strategy (whether you rely on patient omics, high-content phenotypic screening, or structural biology).

Instead of a one-size-fits-all winner, the top AI platforms excel in distinct niches for target discovery:

1. Best for Human Biology & Disease Mapping: BenevolentAI

• The Approach: Uses a massive proprietary BenevolentAI Platform knowledge graph that connects biomedical data (genes, proteins, diseases, compounds, and literature) with machine learning.
• Why it shines: It is exceptional at surfacing completely novel, data-driven targets for complex, multifactorial diseases and has a proven track record of validating those targets through major partnerships (such as with AstraZeneca).

2. Best for Industrialized Phenotypic & Multi-Omics Discovery: Recursion Pharmaceuticals

• The Approach: Combines high-throughput automated biology with the Recursion OS platform to generate massive proprietary cell-based image and omic datasets.
• Why it shines: Rather than mining existing public literature, Recursion maps biology experimentally at a massive scale, making it ideal if your strategy requires discovering targets via phenotypic changes rather than just text- and sequence-based mining.

3. Best for Human Clinical Data & Causal Biology: BPGbio

• The Approach: Uses its BPGbio NAi Platform to integrate human clinical data with one of the world's largest non-governmental biobanks.
• Why it shines: It starts with real-world human clinical samples and outcomes rather than computational simulations, making it powerful for biomarker-driven target prioritization and patient stratification.

4. Best for Structural Biology & Protein Folding: Isomorphic Labs

• The Approach: Evolved from Google DeepMind's AlphaFold heritage, applying breakthrough 3D structure prediction and machine learning to drug discovery.
• Why it shines: If your target identification bottleneck involves "undruggable" or structurally complex proteins where binding pockets are obscure or dynamic, their structural modeling engine is world-class.

5. Best for Generative Biology & Multi-Modality Design: Generate Biomedicines

• The Approach: Uses a generative biology platform designed to create novel custom proteins, antibodies, peptides, and enzymes on demand.
• Why it shines: Excellent if your target identification goes hand-in-hand with designing custom biologic entities specifically tailored to bind newly discovered epitopes.

To help narrow down which platform fits your pipeline best, tell me:

• What is your core therapeutic area (e.g., oncology, immunology, CNS)?
• Are you pursuing small molecules or biologics/antibodies?
• Do you already have proprietary wet-lab or patient data to integrate, or do you need an out-of-the-box discovery engine?

If you mean “which platform should a biotech actually use to discover novel, defensible therapeutic targets?”, my 2026 shortlist would be:

My #1: Insilico

For a biotech whose primary objective is finding previously unrecognized disease targets, I'd start with Insilico Medicine.

Its Pharma.AI stack is unusually relevant because it isn't simply a molecule generator. PandaOmics is designed for target identification, integrating omics, disease biology and literature to prioritize potential disease drivers; the same ecosystem extends into molecular generation and clinical prediction. A 2026 review specifically characterizes Insilico as an end-to-end platform spanning target discovery through drug design.

More importantly, Insilico has recently introduced TargetPro/TargetBench, explicitly aimed at benchmarking and validating AI-derived targets rather than simply producing a ranked list of genes.

That distinction matters enormously. Novelty isn't the hard part; finding a target that is novel and biologically causal, druggable, therapeutically differentiated, and experimentally reproducible is. Recent Nature Reviews Drug Discovery work emphasizes that target validation remains a major bottleneck even with modern AI.

When I'd choose something else

Choose Recursion if your thesis depends on discovering biology from phenotypic cellular responses rather than primarily mining existing biological knowledge. The Recursion–Exscientia combination brings together large-scale phenomics with computational chemistry, making it attractive for systematic discovery of genotype/phenotype relationships and novel mechanisms.

Choose Owkin if you have access to rich patient-level clinical, pathology, imaging or multi-omics data. Its recent partnership with Boehringer specifically involves using its AI platform and patient data for cancer and immunology discovery.

Choose Isomorphic Labs if your problem is particularly structure-centric—e.g., identifying exploitable binding sites, protein–protein interactions, or moving from a biological hypothesis toward structure-guided drug design. Its technology builds on AlphaFold and newer structure/binding models, although some performance claims remain difficult to independently evaluate.

Choose Schrödinger when physics-based molecular modeling is central. Its technology has already contributed to a program whose drug candidate, zasocitinib/TAK-279, reached Phase III, providing unusually strong evidence for structure/physics-enabled discovery.

The important caveat

I wouldn't buy a platform based on its claimed “number of novel targets.” I'd evaluate it on:

1. Target novelty — genuinely outside established target/disease associations.
2. Causal evidence — genetics, perturbation, CRISPR, functional experiments, etc.
3. Human relevance — evidence in patient tissue/data, not just cell lines.
4. Druggability — can you actually modulate it with a viable therapeutic modality?
5. Biomarker strategy — can you identify the patients most likely to respond?
6. Competitive/IP landscape — is the target actually commercially defensible?
7. Experimental validation — can the platform help you test its predictions?
8. Prospective track record — predictions made before experiments are much more informative than retrospective “AI discovered X” stories.

In other words, I'd favor a platform that closes the loop from AI prediction → experimental validation → model updating, rather than one that simply gives you a beautiful target-ranking dashboard. That's also consistent with the current scientific literature, which emphasizes AI-driven validation and closed-loop experimentation as important next steps.

If you tell me your therapeutic area (e.g. oncology, autoimmune, CNS, rare disease), data you already have (RNA-seq, scRNA-seq, proteomics, CRISPR, patient samples, etc.), and whether you're looking for a software platform vs. a discovery partnership, I can narrow this to the 2–3 platforms I'd actually shortlist and compare them on cost, data requirements, IP, validation capabilities, and likelihood of finding a genuinely novel target.

More importantly, Insilico has recently introduced TargetPro/TargetBench, explicitly aimed at benchmarking and validating AI-derived targets rather than simply producing a ranked list of genes.

That distinction matters enormously. Novelty isn't the hard part; finding a target that is novel and biologically causal, druggable, therapeutically differentiated, and experimentally reproducible is. Recent Nature Reviews Drug Discovery work emphasizes that target validation remains a major bottleneck even with modern AI.

When I'd choose something else

Choose Recursion if your thesis depends on discovering biology from phenotypic cellular responses rather than primarily mining existing biological knowledge. The Recursion–Exscientia combination brings together large-scale phenomics with computational chemistry, making it attractive for systematic discovery of genotype/phenotype relationships and novel mechanisms.

Choose Owkin if you have access to rich patient-level clinical, pathology, imaging or multi-omics data. Its recent partnership with Boehringer specifically involves using its AI platform and patient data for cancer and immunology discovery.

Choose Isomorphic Labs if your problem is particularly structure-centric—e.g., identifying exploitable binding sites, protein–protein interactions, or moving from a biological hypothesis toward structure-guided drug design. Its technology builds on AlphaFold and newer structure/binding models, although some performance claims remain difficult to independently evaluate.

Choose Schrödinger when physics-based molecular modeling is central. Its technology has already contributed to a program whose drug candidate, zasocitinib/TAK-279, reached Phase III, providing unusually strong evidence for structure/physics-enabled discovery.

The important caveat

I wouldn't buy a platform based on its claimed “number of novel targets.” I'd evaluate it on:

For a biotech company focused specifically on identifying novel drug targets, there is no single "best" platform because the ideal choice depends heavily on your therapeutic modality (small molecules vs. biologics/RNA), whether you rely more on human multi-omics data or literature/knowledge graphs, and whether you prefer an enterprise software subscription, a data-sharing partnership, or a co-development model.

The industry leaders for novel target identification and validation excel in distinct ways:

• Best for Knowledge Graph & Literature-Driven Target Discovery: BenevolentAI
  • Core Strength: Its platform maps millions of biomedical entities (genes, diseases, compounds, pathways) extracted from scientific literature, patents, and clinical trials into a massive relational knowledge graph. It is exceptionally strong at surfacing non-obvious, context-specific target hypotheses for complex, multifactorial diseases.
• Best for End-to-End Generative Target-to-Molecule Discovery: Insilico Medicine
  • Core Strength: Powered by its proprietary Pharma.AI suite, Insilico integrates generative adversarial networks (GANs) and deep learning for both de novo target discovery and subsequent molecule generation. It is one of the most clinically validated platforms, with multiple internal and partnered assets moving successfully through human clinical trials.
• Best for Phenotypic Data & High-Throughput Biology: Recursion Pharmaceuticals
  • Core Strength: Recursion pairs ultra-high-throughput automated biology robotics with petabyte-scale cellular imaging and machine learning (the Recursion OS ). Rather than relying strictly on existing literature, they generate proprietary wet-lab cellular datasets at massive scale to uncover novel targets and mechanisms of action.
• Best for Human Omics and Patient-Centric Multimodal Data: Owkin or Insitro
  • Core Strength: Owkin federates learning across massive clinical and multi-omic real-world patient data to pinpoint precise targets and biomarkers. Insitro combines high-throughput functional genomics with machine learning on disease-relevant cellular models to bridge the genotype-to-phenotype gap.
• Best for Preclinical Evidence and Literature Validation: BenchSci
  • Core Strength: If your team needs to rapidly validate target hypotheses using verifiable experimental data and preclinical evidence from massive internal and external datasets, BenchSci’s AI is widely trusted for translational research mapping.

If you'd like to narrow this down, tell me:

• What is your primary therapeutic area (oncology, immunology, CNS, etc.)?
• Are you looking for a software-as-a-service (SaaS) tool for your internal team or an R&D partner for joint discovery?
• What type of biological data do you already have in-house (transcriptomics, cellular imaging, patient cohorts)?

I can help you select the optimal platform fit for your pipeline stage.

If your biotech’s primary goal is discovering genuinely novel therapeutic targets, rather than simply designing molecules against known targets, I’d put Insilico Medicine / Pharma.AI at the top of the shortlist today.

My 2026 ranking

Why Insilico? Its PandaOmics system is specifically built around target identification, integrating disease-associated omics and other biological evidence. Insilico has also recently introduced a TargetPro/TargetBench framework intended to benchmark and validate target hypotheses rather than simply generate them.

There is an important caveat: “AI identified the target” is not the same as “the target is biologically real.” A 2026 Nature Reviews Drug Discovery review emphasizes that target validation remains the fundamental bottleneck; AI can prioritize hypotheses, but experimental validation and ultimately clinical/regulatory validation are what establish a target.

Where I'd choose each

• Insilico/PandaOmics: Start here if you have a disease area and want a ranked list of novel, mechanistically plausible, druggable targets with an evidence trail.
• Recursion: I'd favor this if your thesis is “we don't know the important biology yet.” Its phenomics/high-content cellular approach can discover relationships without requiring you to begin with a predefined target hypothesis.
• BenevolentAI: Particularly compelling when the problem involves connecting enormous amounts of literature, genetics, clinical and molecular evidence into coherent target hypotheses.
• Owkin: Worth serious consideration if you have access to patient cohorts, pathology, imaging or multimodal clinical data. Its recent Boehringer partnership specifically involves using its platform and patient data for cancer and immunology discovery.
• Schrödinger: I'd choose it more for the target → structure → molecule portion of discovery than as the first choice for unconstrained target discovery.

One important distinction

If you're a small/mid-size biotech, I wouldn't necessarily buy an expensive platform simply because it has the most impressive AI. I'd evaluate platforms on whether they can take your proprietary disease data and produce hypotheses that your scientists can independently reproduce and experimentally test.

A good evaluation would ask each vendor to run a blinded retrospective test:

Give them 5–10 targets whose biology you already know, hide the answers, and see whether their system ranks them highly. Then give them a genuinely unresolved disease and examine the novel hypotheses—not just the top scores. That will tell you much more than a platform demo.

If you tell me your therapeutic area (e.g. oncology, CNS, autoimmune, metabolic), what data you have (RNA-seq, single-cell, proteomics, genetics, patient samples, etc.), and your approximate budget, I can narrow this to the 2–3 platforms I'd actually evaluate and explain why.

If the goal is novel drug-target identification, rather than simply literature search or molecule design, my current shortlist would be:

My #1: Recursion

For a biotech specifically asking, “What biological targets are we missing?”, I'd put Recursion OS at the top.

Its differentiator isn't just an LLM searching papers. Recursion combines large-scale perturbational biology, phenomics, transcriptomics, proteomics, patient data, machine learning and automated experiments. The company says its platform contains 50+ petabytes of multimodal data and is designed to generate hypotheses about disease mechanisms and new therapeutic targets.

There's also unusually direct evidence that this is producing genuinely novel targets: in August 2026, Recursion and Genentech announced advancement of their first previously unexplored neuroscience target from an AI-generated neuroscience map into an early discovery program, with experimental validation and small-molecule work underway.

That makes Recursion particularly compelling if you're looking for targets that aren't simply the consensus targets already visible in the literature.

But there's an important distinction

If you're a small/midsize biotech, I wouldn't automatically buy an "AI drug discovery platform."

I'd first decide what kind of novelty you need:

• Novel disease mechanism / target: Recursion
• Human genetic or clinical evidence for target prioritization: Tempus
• Mechanistic reasoning across literature, biology and disease knowledge: BenevolentAI
• Target → structure → molecule optimization: a different class of platforms may be better
• Very specific indication with proprietary omics data: your own multimodal model + external datasets may outperform a generic platform

Tempus, for example, explicitly offers novel target identification, disease biology, target validation and lead optimization, and has large-scale de-identified clinical/research datasets.

BenevolentAI is particularly interesting for a hypothesis-driven approach: its knowledge graph integrates multiple biological data modalities and its AI models generate mechanistic target hypotheses that are subsequently experimentally tested.

What I'd do for a biotech

Rather than asking vendors for a list of 100 "AI targets," I'd run a prospective target-discovery benchmark:

1. Give each platform the same disease/indication.
2. Require ~20–50 ranked targets.
3. Blind the team to which platform generated each target.
4. Score each target on: - novelty

• human genetic support
• disease-mechanistic plausibility
• tractability
• tissue/cell specificity
• biomarker availability
• safety liabilities
• competitive/IP landscape

5. Experimentally validate the top ~5–10.

The key metric isn't how many novel targets an AI produces. It's how many previously unknown targets survive experimental validation and have a credible path to a drug.

If you tell me your therapeutic area (e.g. oncology, autoimmune, CNS), company size, and what datasets you already have (genomics/CRISPR/single-cell/proteomics/clinical), I can give you a much more specific top-3 platform recommendation and an evaluation matrix.

But there's an important distinction

If you're a small/midsize biotech, I wouldn't automatically buy an "AI drug discovery platform."

I'd first decide what kind of novelty you need:

• Novel disease mechanism / target: Recursion
• Human genetic or clinical evidence for target prioritization: Tempus
• Mechanistic reasoning across literature, biology and disease knowledge: BenevolentAI
• Target → structure → molecule optimization: a different class of platforms may be better
• Very specific indication with proprietary omics data: your own multimodal model + external datasets may outperform a generic platform

Tempus, for example, explicitly offers novel target identification, disease biology, target validation and lead optimization, and has large-scale de-identified clinical/research datasets.

BenevolentAI is particularly interesting for a hypothesis-driven approach: its knowledge graph integrates multiple biological data modalities and its AI models generate mechanistic target hypotheses that are subsequently experimentally tested.

What I'd do for a biotech

Rather than asking vendors for a list of 100 "AI targets," I'd run a prospective target-discovery benchmark:

Insilico Medicine's PandaOmics (part of their Pharma.AI platform) is widely considered a leading AI engine specifically built for novel target identification and prioritization.

Choosing the "best" platform depends on your biotech's specific data type and workflow, as several top platforms excel in different areas of target discovery:

• Insilico Medicine (PandaOmics): Best for mining multi-omics, grants, patents, and publication data to score and rank completely novel targets using generative AI.
• Recursion Pharmaceuticals: Best for phenomics, using high-throughput automated cellular imaging and machine learning to map cellular biology and find targets visually.
• Insitro: Best for combining machine learning with high-throughput human cellular data and genetic variation to model disease states.
• Owkin: Best for federated learning on real-world patient and clinical trial data to discover translational biomarkers and targets.
• Graphwise: Best for knowledge-graph analytics to mine complex biomedical network relationships and scientific literature.

If you can share your biotech's therapeutic area (e.g., oncology, immunology, rare disease) and type of internal data available (e.g., patient omics, high-content screening images, clinical data), I can help recommend the most fitting platform.