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 the optimal choice depends entirely on your biotech’s specific modality (small molecules vs. biologics/proteins), your data maturity, and whether you need an end-to-end proprietary pipeline or an enterprise software tool to empower your internal wet-lab scientists.

Leading AI platforms for target identification and biological discovery excel in distinct areas:

• Isomorphic Labs: Born from DeepMind (the creators of AlphaFold), Isomorphic Labs focuses on deep integration of next-generation AI and physical modeling to understand biological mechanisms at an atomic level and rationally design therapeutics . Best for cutting-edge target-to-structure understanding.
• BenevolentAI: Features the proprietary Benevolent Platform, which constructs a massive, interconnected biomedical knowledge graph combining literature, patents, genetics, and multi-omics data . It is specifically built to surface and validate novel disease targets hidden within complex data networks.
• Insilico Medicine: Utilizes the end-to-end Pharma.AI platform, which integrates generative AI and reinforcement learning explicitly for target discovery and novel molecule generation . Proven track record of taking AI-discovered targets and custom molecules into clinical trials.
• BenchSci: Excellent for early-stage landscape mapping. Their AI platform (Mandala ) uses biomedical computer vision and NLP to mine published literature and experimental data, helping researchers instantly see what has been tested and validate whether a target is worth pursuing before running costly assays.
• Tempus AI: Unrivaled if your target discovery strategy relies on real-world clinical and multi-modal patient data (especially in oncology and precision medicine), allowing biotechs to correlate patient outcomes with genomic profiles to find novel targets.

To help narrow down the ideal platform for your team, tell me:

• What is your therapeutic area (e.g., oncology, immunology, CNS)?
• Are you looking for a software-as-a-service/licensing partner or a co-development collaboration?
• Do you primarily work with small molecules or biologics?

There is no single “best” AI platform for identifying novel drug targets; the best choice depends on whether your biotech is trying to discover new biology, prioritize genetically supported targets, find mechanisms from phenotypic data, or build an end-to-end AI drug discovery engine. The leading options differ substantially.

Short answer

For a biotech focused specifically on novel target identification, my shortlist would be:

ScienceDirect PubMed Central (PMC)## If I were advising a biotech choosing today

1. For a small biotech with a novel disease area: Insilico Medicine

This is probably the closest fit if your goal is:

“Find an interesting target → understand why it matters → design a drug against it.” Its PandaOmics platform combines biological data, multi-omics, and literature-derived evidence for target identification, while its broader Pharma.AI stack connects target discovery to molecule design. Insilico has also published work on AI-driven target identification approaches and has advanced AI-derived programs clinically.

Best for:

• rare disease
• fibrosis
• oncology
• immunology
• targets where you have human datasets but unclear mechanisms

Main caveat: You still need experimental validation. AI ranking is not proof of causality.

---

2. For discovering biology humans have missed: Recursion

Recursion takes a different philosophy: instead of starting from a target hypothesis, it generates large-scale maps of cellular states and perturbations, then uses AI to identify disease mechanisms and drug opportunities.

Best for:

• oncology
• rare diseases
• poorly understood pathways
• phenotype-first discovery

Main caveat: It is less of a “give me the top 20 genetically validated targets” engine and more of a biology discovery machine.

---

3. For mechanism mining and hypothesis generation: BenevolentAI

BenevolentAI’s approach is based heavily on biomedical knowledge graphs: connecting publications, clinical information, pathways, genes, and compounds to generate hypotheses.

Best for:

• diseases with enormous literature complexity
• repurposing opportunities
• finding non-obvious biological connections

Main caveat: Knowledge graphs can surface correlations that still require strong experimental follow-up.

---

My practical recommendation by biotech stage

Seed / Series A biotech

I would usually combine:

1. Open Targets → establish genetic evidence
2. Insilico PandaOmics or BenevolentAI → generate ranked hypotheses
3. Internal wet-lab validation → CRISPR, organoids, animal models, biomarkers

This reduces the risk of trusting a single AI “black box.”

Platform biotech building a discovery engine

Consider:

• Recursion for proprietary biological datasets
• Insilico for integrated target-to-drug workflows
• Schrödinger for structure/chemistry optimization

Pharma partnership strategy

A platform with:

• proprietary data,
• validated experimental loops,
• clinical-stage assets,
• and reproducible target-selection methods

will generally be more valuable than a model that only produces target lists.

My ranking for novel target discovery specifically

1. Insilico Medicine (PandaOmics) — strongest overall target-to-drug integration
2. Recursion — strongest for discovering unexpected biology
3. BenevolentAI — strongest for literature/knowledge-driven hypotheses
4. Open Targets — best independent validation layer
5. Schrödinger — excellent downstream, less focused on discovering unknown targets

The most successful biotech teams will likely not pick one platform; they will use AI platforms as a target hypothesis engine plus evidence filter, then differentiate themselves through proprietary biology and experimental validation.

If you tell me the biotech’s therapeutic area (oncology, CNS, immunology, rare disease, etc.), data assets (omics, CRISPR, patient samples), and whether you want a vendor tool vs a strategic partnership, I can narrow this to the best fit.

If your goal is specifically discovering novel, defensible drug targets—rather than molecule design—I’d put Recursion at the top of the shortlist in 2026, but the “best” platform depends heavily on what data your biotech has and whether you want software, a partnership, or an end-to-end discovery engine.

My ranking

1. Recursion — my first choice for novel targets

Recursion is unusually strong if you're trying to find targets that aren't already obvious from the literature. Its approach combines large-scale phenomics, transcriptomics, proteomics, chemical data and patient data into biological maps, then uses ML to identify previously unrecognized relationships. Its platform explicitly spans target identification through drug development.

The important recent validation: in August 2026, Recursion and Genentech announced that they had identified and biologically validated a previously unexplored neuroscience target, which was then advanced into small-molecule hit generation.

That's particularly relevant to your question: it's evidence of the workflow going AI → novel target → experimental validation → drug program, rather than merely predicting known targets.

2. Insilico — best if you want target → molecule in one system

Insilico is attractive when your objective isn't just “find an interesting target” but “find a target and rapidly turn it into a drug program.”

Its Pharma.AI stack combines target discovery, biology, chemistry and generative design. It also has one of the stronger clinical validation stories among AI-native drug-discovery companies. Recent industry assessments put Insilico among the leaders in clinical validation.

I'd favor it over Recursion if your biotech has a small discovery team and wants an external partner to carry much of the computational discovery burden.

3. Isomorphic Labs — best for structure/mechanism-heavy biology

If the target is a protein where 3D structure, protein–ligand interactions, complex formation or mechanistic understanding are central, Isomorphic Labs is extremely compelling.

But there's an important distinction: Isomorphic is primarily a drug-discovery partner, not a conventional SaaS platform you simply subscribe to. Its AlphaFold-derived technology is much more relevant to understanding and designing against targets than to doing unbiased phenotypic target discovery across an entire disease biology.

4. BenevolentAI — particularly interesting for hypothesis generation

For a biotech sitting on large amounts of omics, genetics, literature and clinical data, knowledge-graph approaches can be powerful. The broader research literature increasingly supports combining knowledge graphs with ML for target prioritization and mechanistic reasoning.

This category is particularly useful for questions like:

“What biological node connects these disease phenotypes, genetic associations and pathways—and is there evidence that perturbing it could be therapeutic?” That's somewhat different from Recursion's experimentally driven phenotype-first approach.

What I'd actually do for a biotech

I wouldn't buy one AI platform and ask it to tell you your next drug target.

I'd build a target-discovery stack with three layers:

1. Discovery layer: Use Recursion/phenomics, genetics, single-cell data, patient data, perturbational data and literature/knowledge graphs to generate hypotheses.

2. Target-assessment layer: For each candidate, score:

• human genetic evidence
• disease/phenotype causality
• tissue/cell-type specificity
• mechanistic plausibility
• biomarker availability
• tractability/druggability
• safety liabilities
• competitive/IP landscape
• evidence that the target is genuinely novel
• evidence that modulation should produce the desired therapeutic phenotype

3. Experimental validation: The AI's output should be a ranked, testable target hypothesis, not the final answer. A 2026 Nature Reviews Drug Discovery review emphasizes that target identification and validation remain major bottlenecks; computational evidence doesn't substitute for biological validation.

My recommendation by biotech situation

• You have little proprietary biological data: → Insilico or BenevolentAI
• You have substantial phenotypic/omics data and want genuinely new biology: → Recursion
• Your target is structurally/mechanistically challenging: → Isomorphic Labs
• You have an internal computational biology team and want control: → build around knowledge graphs + your own models, potentially incorporating Schrödinger/NVIDIA tooling
• You want one partner to go from target discovery through candidate generation: → Recursion or Insilico

If you tell me your therapeutic area (e.g. oncology, CNS, autoimmune, metabolic), what data you have (genomics/single-cell/proteomics/patient data/CRISPR/phenotypic screens), and whether you're looking for a SaaS tool or a strategic partner, I can narrow this to the 2–3 platforms I'd actually contact and explain why.

The important recent validation: in August 2026, Recursion and Genentech announced that they had identified and biologically validated a previously unexplored neuroscience target, which was then advanced into small-molecule hit generation.

That's particularly relevant to your question: it's evidence of the workflow going AI → novel target → experimental validation → drug program, rather than merely predicting known targets.

2. Insilico — best if you want target → molecule in one system

Insilico is attractive when your objective isn't just “find an interesting target” but “find a target and rapidly turn it into a drug program.”

Its Pharma.AI stack combines target discovery, biology, chemistry and generative design. It also has one of the stronger clinical validation stories among AI-native drug-discovery companies. Recent industry assessments put Insilico among the leaders in clinical validation.

I'd favor it over Recursion if your biotech has a small discovery team and wants an external partner to carry much of the computational discovery burden.

3. Isomorphic Labs — best for structure/mechanism-heavy biology

If the target is a protein where 3D structure, protein–ligand interactions, complex formation or mechanistic understanding are central, Isomorphic Labs is extremely compelling.

But there's an important distinction: Isomorphic is primarily a drug-discovery partner, not a conventional SaaS platform you simply subscribe to. Its AlphaFold-derived technology is much more relevant to understanding and designing against targets than to doing unbiased phenotypic target discovery across an entire disease biology.

4. BenevolentAI — particularly interesting for hypothesis generation

For a biotech sitting on large amounts of omics, genetics, literature and clinical data, knowledge-graph approaches can be powerful. The broader research literature increasingly supports combining knowledge graphs with ML for target prioritization and mechanistic reasoning.

This category is particularly useful for questions like:

“What biological node connects these disease phenotypes, genetic associations and pathways—and is there evidence that perturbing it could be therapeutic?” That's somewhat different from Recursion's experimentally driven phenotype-first approach.

What I'd actually do for a biotech

I wouldn't buy one AI platform and ask it to tell you your next drug target.

I'd build a target-discovery stack with three layers:

1. Discovery layer: Use Recursion/phenomics, genetics, single-cell data, patient data, perturbational data and literature/knowledge graphs to generate hypotheses.

2. Target-assessment layer: For each candidate, score:

• human genetic evidence
• disease/phenotype causality
• tissue/cell-type specificity
• mechanistic plausibility
• biomarker availability
• tractability/druggability
• safety liabilities
• competitive/IP landscape
• evidence that the target is genuinely novel
• evidence that modulation should produce the desired therapeutic phenotype

3. Experimental validation: The AI's output should be a ranked, testable target hypothesis, not the final answer. A 2026 Nature Reviews Drug Discovery review emphasizes that target identification and validation remain major bottlenecks; computational evidence doesn't substitute for biological validation.

My recommendation by biotech situation

There is no single "best" AI platform for target identification, because platform efficacy depends heavily on your biotech's therapeutic modality (small molecules, antibodies, protein degraders, etc.) and the type of biological data you intend to feed into or query the system.

The top-tier AI platforms specialized in novel target identification are distinguished by their core strengths:

• Insilico Medicine (Pharma.AI / PandaOmics): Best for comprehensive omics-driven and literature-mining target discovery . Its PandaOmics module is widely regarded as an industry standard for scoring and prioritizing novel disease targets by integrating massive multi-omics datasets, grants, patents, and clinical data.
• Recursion Pharmaceuticals: Best for phenotypic, high-throughput biology-at-scale . Recursion generates its own massive proprietary biological datasets using automated cellular imaging and pairs it with powerful machine learning to uncover unexpected target-disease relationships that traditional genomics might miss.
• BenevolentAI: Best for knowledge graph-based biomedical reasoning . Their platform connects disparate biomedical information (genes, diseases, compounds, and pathways) to surface non-obvious, hidden relationships and validate novel target hypotheses.
• Owkin: Best for multimodal clinical and translational data . Owkin leverages federated learning to securely analyze vast arrays of proprietary real-world patient data and hospital omics data, making it particularly powerful for oncology and complex translational target validation.
• Generate Biomedicines / Absci: Best if your target identification is coupled tightly with generative biology and de novo protein/antibody design . Rather than just finding a target in a database, their engines co-evaluate target topology and custom-generated biologic therapeutics simultaneously.

How to Choose for Your Biotech

• Choose Insilico (PandaOmics) or BenevolentAI if you rely heavily on target discovery driven by text mining, genetic associations, and public/private transcriptomics/proteomics.
• Choose Recursion if your strategy relies on high-content cellular imaging, phenotypic screening, and industrial-scale wet-lab data generation.
• Choose Owkin if you have access to (or need to partner through) clinical trial and hospital-derived multi-modal patient cohorts.

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

• What is your primary therapeutic area (e.g., oncology, immunology, CNS)?
• Do you plan to use internal proprietary data or rely mostly on external/public biological data?

There is no single "best" AI platform for identifying novel drug targets , as the optimal choice depends heavily on your biotech’s specific modality (small molecules, antibodies, or genetic medicines), your internal data assets, and whether you prefer an omics-mining approach or a phenotypic screening approach.

Leading AI platforms excel in distinct niches for upstream target discovery:

• Insilico Medicine (Pharma.AI / PandaOmics): Best for multi-omics data integration and literature mining. PandaOmics uses generative AI and massive knowledge graphs to score and prioritize disease targets by analyzing transcriptomics, proteomics, clinical data, and millions of scientific papers. It is a premier choice if your strategy relies on finding first-in-class targets backed by deep human genetics and bioinformatics.
• Recursion Pharmaceuticals (Recursion OS): Best for high-throughput phenomics and cellular imaging. Recursion maps biology by coupling automated, high-content cellular imaging with deep learning. Rather than starting with a known protein or gene, their platform identifies novel targets and mechanisms of action by observing microscopic phenotypic changes across thousands of cellular perturbations.
• Insitro: Best for combining machine learning with proprietary functional genomics. Insitro builds predictive machine learning models using human genetic data coupled with engineered cellular disease models. They specialize in generating proprietary, high-quality in vitro data to discover targets that are genuinely disease-causal rather than correlative.
• BenevolentAI (Benevolent Platform): Best for biomedical knowledge-graph traversal and target validation. This platform excels at connecting fragmented biomedical information—spanning patents, molecular data, clinical trials, and omics data—to surface hidden biological relationships and novel disease targets.

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

• What is your primary therapeutic area (e.g., oncology, immunology, CNS)?
• Do you plan to rely mostly on public/existing datasets or generate proprietary wet-lab data to feed the platform?

If you mean “which platform should a biotech actually license/use to discover novel targets, rather than simply search literature?”, my current shortlist would be:

My #1: Insilico/PandaOmics

For a typical small-to-mid-size biotech that wants to identify and rank novel targets, I'd start with PandaOmics. It combines multi-omics, genetics, disease biology and literature/knowledge-graph information for target identification and prioritization. Reviews of current AI drug-discovery platforms specifically identify PandaOmics as a target-identification platform, alongside Recursion's phenomics approach and other major systems.

The important distinction is that target discovery isn't the same as molecule generation. PandaOmics is particularly relevant if your question is:

“Given disease X, what previously underappreciated genes/proteins have convergent evidence that makes them worth experimentally validating?”

But Recursion may be better for truly novel biology

If your biotech has the resources to work with a much broader experimental platform, I'd seriously consider Recursion. Its advantage is the proprietary experimental dataset: the company says its platform contains >50 PB spanning phenomics, transcriptomics, proteomics, ADME and patient data, with automated experiments feeding results back into its models.

There's also a particularly relevant recent validation: in August 2026, Recursion announced that its collaboration with Genentech had produced its first neuroscience target, which was experimentally validated and moved toward small-molecule hit generation.

That makes Recursion unusually interesting if your goal isn't merely ranking known biological hypotheses, but discovering targets from perturbational/phenotypic biology.

Don't underestimate knowledge-graph platforms

For a biotech with a strong computational biology team, I'd also evaluate Standigm ASK and newer knowledge-graph systems. Standigm's approach integrates heterogeneous biological data into a KG, ranks targets and exposes the evidence behind predictions. Its published IPF work actually experimentally validated several predicted targets.

This is important because “AI says target X” isn't sufficient for a drug program. You want an auditable chain:

disease → genetics → cell state → pathway → target → mechanism → druggability → safety → biomarker → experiment

Recent 2026 research is moving in exactly this direction: TargetMap, for example, uses LLM-based reasoning over complete biological pathways rather than relying only on local graph relationships.

What I'd choose by biotech situation

If I were building a new biotech today:

• Limited wet-lab resources / need a target shortlist: PandaOmics
• Deep proprietary biology + screening capability: Recursion
• Rare disease / oncology / genetics-heavy program: PandaOmics + specialist human-genetics datasets
• Mechanism-heavy biology: Standigm/Plex/knowledge-graph approach
• Computationally sophisticated team: build an internal evidence graph and use commercial AI platforms as inputs rather than relying on one black box
• Want target → molecule → preclinical in one ecosystem: Insilico or Recursion become especially attractive

One caveat: there is no platform I'd trust to declare a target “validated.” The 2026 Nature Reviews Drug Discovery review emphasizes that target identification remains highly uncertain and that definitive validation ultimately comes through successful drug development.

If you tell me your therapeutic area (e.g. oncology, CNS, autoimmune, rare disease), company size, and whether you have your own omics/screening data, I can narrow this to the 2–3 platforms I'd actually evaluate and compare them on target novelty, evidence quality, proprietary data, cost, and track record.

There is no single "best" AI platform for identifying novel drug targets , as optimal selection depends heavily on your biotech’s specific modality, therapeutic area, and internal wet-lab infrastructure . Instead, the market is led by distinct platform archetypes tailored to different discovery strategies.

The top AI platforms for novel target identification are categorized below by their core strengths:

• Insilico Medicine (PandaOmics) : Best for comprehensive omics and literature mining . Its PandaOmics module integrates massive multi-omics datasets (transcriptomics, proteomics, genomics) alongside millions of data points from grants, patents, and publications to score and prioritize novel disease-driving targets . It represents one of the most clinically validated end-to-end engines.
• Recursion Pharmaceuticals : Best for phenomics-first and high-content screening . Recursion runs massive automated wet-lab infrastructure that generates petabytes of biological image data , training foundational models (such as Phenom on NVIDIA's BioNeMo ) to flag dysregulated cellular biology and uncover overlooked therapeutic hypotheses without needing a predefined target hypothesis.
• Isomorphic Labs (Google DeepMind spinout) : Best for structural biology and deep-learning prediction . Building on advanced protein-folding and predictive architecture, Isomorphic’s design engine maps complex binding sites and non-obvious intervention points , backed by major enterprise partnerships with Novartis and Eli Lilly.
• BenevolentAI : Best for biomedical knowledge graphs . Their platform structures fragmented biomedical data into a massive relational graph, allowing translational scientists to reason over hidden connections between genes, diseases, and compounds for target discovery and drug repurposing.
• insitro : Best for translational and clinical data integration . Founded heavily on generating proprietary cellular, clinical, and genetic data via automated engineering, insitro excels at bridging human patient data with predictive disease models.

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

• What is your therapeutic focus (oncology, CNS, rare disease, immunology)?
• Are you looking for a software-as-a-service (SaaS) license or a co-development partnership?

If your biotech’s specific goal is novel target identification, rather than molecule design, I’d put Insilico Medicine’s Pharma.AI at the top of the shortlist today—especially its target-discovery stack. Its PandaOmics/TargetPro approach is explicitly aimed at finding and prioritizing disease-relevant targets, and Insilico has recently published a target-identification benchmarking framework.

My shortlist

The distinction matters. A 2026 review of the field characterizes Recursion and BenevolentAI as particularly strong in elucidating biology and generating therapeutic hypotheses, while Insilico is notable for its integrated target-to-drug workflow.

If I were choosing for a biotech

1. Insilico — strongest default choice.
I'd choose it if your workflow is something like:

disease → multimodal data → novel target hypotheses → target ranking → tractability → biomarker strategy → molecule design

Its advantage is that target discovery isn't isolated from downstream druggability. The platform is designed to connect target identification with generative chemistry and clinical prediction.

2. Recursion — potentially better for genuinely unexplored biology.
Recursion takes a somewhat different approach: large-scale phenotypic experiments, imaging, omics and machine learning are combined to construct "maps of biology," rather than starting primarily from known disease hypotheses. That can be powerful when you're trying to discover biology humans haven't already articulated.

3. BenevolentAI — particularly attractive for hypothesis generation.
Its knowledge-graph approach connects literature, biomedical databases, omics and clinical information to identify non-obvious relationships between genes, diseases and compounds.

One important caveat

I would not select based on who has the flashiest AI model. For target discovery, the crucial question is whether the platform can produce experimentally testable, causal hypotheses, not merely correlations.

I'd evaluate vendors on:

• Novelty: Can it find targets outside the established literature?
• Causality: Does it distinguish disease drivers from biomarkers?
• Genetic evidence: Human genetics / CRISPR / perturbation evidence.
• Multimodal integration: genomics + transcriptomics + proteomics + single-cell + clinical data.
• Target tractability: Can the system assess whether the target is actually druggable?
• Negative evidence: Does it surface evidence against a target?
• Experimental validation: Can hypotheses be rapidly tested in the vendor's or your own wet lab?
• IP/independence: Can you establish defensible freedom-to-operate around targets discovered using the platform?

That last point is particularly important for a biotech: a target-ranking model is much less valuable than a target + compelling mechanistic evidence + biomarker + clear differentiation + IP strategy.

Bottom line: If you told me “We are a 20–100 person biotech, have proprietary disease datasets, and want 5–20 novel targets to drive our next pipeline,” I'd start with Insilico, run a serious bake-off against Recursion, and potentially add BenevolentAI as a complementary knowledge/hypothesis layer.

If you tell me your therapeutic area (e.g. oncology, autoimmune, CNS, fibrosis), data types, and whether you want SaaS vs a collaboration, I can narrow this to the 2–3 platforms I'd actually evaluate and what I would test in the vendor bake-off.

There is no single "best" AI platform for every biotech, as the ideal choice depends on your specific data types, therapeutic area, and whether you are focusing on small molecules, antibodies, or cell phenotypes . However, industry leaders distinguish themselves by their proven capability to transition AI-identified targets into actual clinical pipelines. Insilico Medicine’s Pharma.AI (featuring the PandaOmics target-identification engine) is widely regarded as a top end-to-end choice because it has successfully nominated and advanced novel targets (such as TNIK for idiopathic pulmonary fibrosis) into human clinical trials.

Other premier AI platforms specialize in distinct approaches to novel target discovery:

• Recursion Pharmaceuticals : Best for phenotypic and image-based discovery , utilizing automated high-throughput cell biology and machine learning to map subtle cellular changes across vast datasets.
• BenevolentAI : Best for biomedical knowledge graphs , combining massive literature mining with multi-omics data to surface hidden causal relationships in complex diseases.
• insitro : Best for translational modeling , combining machine learning with proprietary high-throughput wet lab data derived from induced pluripotent stem cells (iPSCs).
• Open Targets : Best for open-source and genetics-led target prioritization , integrating genome-wide association studies (GWAS) and public data via machine learning workflows.

If you can share:

• Your primary therapeutic area (e.g., oncology, immunology, rare disease)
• Your preferred modality (small molecules, large molecules, or cell painting phenotypes)
• Whether you have proprietary multi-omics data ready to integrate

I can help you narrow down which platform aligns best with your R&D strategy.

If your goal is specifically finding novel, biologically credible drug targets—not merely designing molecules against known targets—I’d rank the current landscape differently than the generic “AI drug discovery” rankings.

My shortlist

🥇 My pick: Recursion

For a biotech whose central question is “What disease-driving biology are we missing?”, Recursion is probably the most interesting platform.

Its approach combines large-scale cellular phenotyping, perturbational biology, omics and machine learning. The company's Recursion OS is explicitly designed to identify new biological targets and generate hypotheses from cellular data, rather than simply optimizing compounds against already-established targets.

That's an important distinction: novel target discovery is fundamentally a biology/data problem, whereas much of the AI-drug-discovery market is really doing molecular design.

🥈 BenevolentAI is particularly interesting for target hypotheses

If your starting point is something like:

“We have a disease area and 20,000 genes/proteins/biomarkers. Find connections that humans haven't noticed.”

I'd seriously investigate BenevolentAI.

Its knowledge-graph approach integrates literature, biomedical databases, omics and clinical information to identify non-obvious relationships between genes, diseases and compounds.

That's complementary to Recursion rather than simply inferior to it.

🥉 Insilico if you want target → drug

Insilico is attractive if you don't want to stop at target identification. Its platform spans target discovery, molecule generation and development, so it can potentially take a hypothesis considerably further down the pipeline.

Where Isomorphic fits

I'd put Isomorphic Labs extremely high if your problem is:

“We've identified a target. Can AI help us understand the protein and design molecules against it?”

Its AlphaFold-derived technology and newer proprietary systems are heavily oriented toward molecular structure and drug design. Isomorphic raised $2.1B in 2026 to scale this approach and move programs toward clinical trials.

But I wouldn't automatically choose it as the first platform for an unexplored disease biology problem.

---

The bigger point: don't buy “AI target discovery” as a black box

For a serious biotech, I'd evaluate platforms on whether they can produce a testable causal hypothesis, not whether they can generate 10,000 targets.

I'd want the platform to give you:

1. Novelty — why isn't this already an established target?
2. Causality — evidence that perturbing it changes disease biology.
3. Human genetics — ideally genetic support in patients.
4. Disease-context specificity — which cell type/state actually matters?
5. Drugability — can you realistically intervene on it?
6. Biomarker strategy — how will you know the target is engaged?
7. Safety liabilities — what happens when you inhibit/activate it systemically?
8. Independent validation — can the hypothesis be tested with CRISPR, organoids, primary cells, animal models, etc.?

That last point is critical. AI can dramatically improve hypothesis generation and prioritization, but experimental biology remains the arbiter of whether a proposed target actually matters.

If I were building a biotech today

I'd actually consider a stack rather than one platform:

Patient genetics + multimodal omics → Recursion/Benevolent-style target hypothesis generation → structural AI (AlphaFold/Isomorphic-type tools) → CRISPR/perturbation validation → medicinal chemistry/AI molecule design.

That gives you substantially more defensible target discovery than asking one foundation model, “What are the next 50 drug targets?”

If you tell me your therapeutic area (e.g., oncology, autoimmune, CNS, metabolic, rare disease) and whether you're looking for small-molecule, antibody, degrader, or RNA targets, I can narrow this to the 3 platforms I'd actually approach—and compare their data access, validation evidence, partnerships, and likely cost.