AI in drug discovery

AI in drug discovery (also called AI drug discovery) is the application of artificial intelligence, machine learning, deep learning, generative models, and natural language processing across the pharmaceutical pipeline to identify drug targets, design new molecules, predict protein structures, screen compound libraries, and optimize clinical trials. As of early 2026, more than 173 AI-discovered drug programs are in clinical development (roughly 94 in Phase I, 56 in Phase II, and 15 in Phase III), and the first analysis of these programs found that AI-discovered molecules cleared Phase I trials at an 80% to 90% rate versus a historical industry average near 40% to 65%.^[1][2] No AI-discovered drug has yet received regulatory approval, and the same analysis found that AI-discovered candidates succeed in Phase II at only about 40%, in line with traditional drugs, leaving open the central debate over whether AI improves success at every clinical stage or mainly at the earliest one.^[1]

The field combines several techniques pioneered by AI for science: AlphaFold for protein structure prediction, generative models for de novo molecule design, machine-learning virtual screening, and AI-assisted clinical-trial design. Leading companies include Isomorphic Labs (a Google DeepMind spinout), Insilico Medicine, Recursion Pharmaceuticals, Schrodinger, and Relay Therapeutics. Two events marked the field's recent milestones: in October 2024 the developers of AlphaFold shared the Nobel Prize in Chemistry, and in June 2025 Insilico Medicine published in Nature Medicine what it described as the first proof-of-concept clinical validation of a drug whose target and molecule were both discovered with generative AI.^[3][4]

What is the drug discovery pipeline and where does AI help?

Traditional drug discovery is a long, expensive, and failure-prone process. Developing a single new drug from initial concept to regulatory approval typically takes 10 to 15 years and costs an estimated $2.6 billion on average, according to the Tufts Center for the Study of Drug Development, a figure that includes the cost of failed programs.^[5][6] Approximately 90% of drug candidates that enter clinical trials fail before reaching approval, with the most common reasons being lack of clinical efficacy (40% to 50%), unmanageable toxicity (30%), poor drug-like properties (10% to 15%), and lack of commercial viability (10%).^[7]

The drug development pipeline consists of several major stages:

Attrition is concentrated in the clinic. In the Biotechnology Innovation Organization's analysis of clinical programs from 2011 to 2020, 52.0% of drugs moved from Phase I to Phase II, 28.9% from Phase II to Phase III, and 57.8% from Phase III to a regulatory submission, and 90.6% of submissions were approved, giving an overall likelihood of approval from Phase I of only 7.9%.^[8] Success rates also vary by therapeutic area: oncology drugs, for example, have historically had lower success rates than many other categories.^[8] This high attrition, combined with the long timelines involved, is the core problem that AI seeks to address.

AI and machine learning can contribute to nearly every stage of this pipeline. In target identification, AI analyzes genomic and proteomic data to pinpoint disease-relevant proteins. In molecule generation and lead optimization, generative models design novel compounds with desired properties. In preclinical development, AI predicts absorption, distribution, metabolism, excretion, and toxicity (ADMET) profiles. In clinical trials, AI helps with patient recruitment, adaptive trial design, and endpoint prediction. Industry analysts estimate that the cumulative effect could cut R&D costs by 25% to 40% and shorten development timelines by several years.^[9]

The main applications, by stage, are summarized below.^[10]

Target identification and validation

Target identification is the first and arguably the most consequential step in drug discovery. Choosing the wrong target leads to years of wasted effort and resources. AI helps by mining vast biological datasets to uncover novel therapeutic targets that would be difficult or impossible to identify through manual analysis alone.

Genomics and multi-omics integration

Modern target discovery relies on the integration of multiple types of biological data, including genomics, transcriptomics, proteomics, and metabolomics. Machine learning models can analyze these multi-omics datasets to identify genes and proteins that play causal roles in disease processes. For example, large language models trained on biomedical literature can systematically analyze disease-associated biological pathways and surface potential targets by connecting genes, diseases, and compounds in knowledge graphs.

Insilico Medicine developed a target identification platform called PandaOmics, which integrates 22 multi-modal data sources to build disease-specific models. Using PandaOmics, Insilico identified TNIK (Traf2- and Nck-interacting kinase) as a novel regulator of lung fibrosis pathways, a target that had not been previously explored for idiopathic pulmonary fibrosis (IPF).^[11] This AI-identified target became the basis for rentosertib, the first AI-designed drug to reach Phase II clinical trials.^[4]

Structure-based design, which depends on knowing the three-dimensional shape of a target protein, was transformed by AlphaFold; that history is covered in its own section below.

How has AlphaFold changed drug discovery?

AlphaFold, developed by Google DeepMind, is one of the most significant breakthroughs in computational biology. Before AlphaFold, experimental determination of protein structures was a bottleneck for structure-based drug design: obtaining a single structure through methods such as X-ray crystallography could take months or years, only a fraction of human proteins had experimentally resolved structures, and many disease-relevant targets lacked structural information entirely.^[12]

The impact was formally recognized in October 2024, when Demis Hassabis and John Jumper were awarded the Nobel Prize in Chemistry for protein structure prediction, sharing the prize with David Baker for computational protein design.^[13][3] By the time of the award, AlphaFold had been used by more than two million researchers across 190 countries.^[3]

AlphaFold 2 and the structure database

AlphaFold 2, announced in late 2020, addressed the long-standing "protein folding problem" by accurately predicting three-dimensional protein structures from amino acid sequences alone. In the CASP14 (Critical Assessment of protein Structure Prediction) competition, it achieved a median GDT (Global Distance Test) score of 92.4, far ahead of all other methods. The AlphaFold Protein Structure Database, launched in July 2021, was expanded in 2022 to cover nearly all catalogued proteins and now contains predicted structures for over 200 million proteins from more than one million species; AlphaFold can produce a prediction in minutes.^[12]

AlphaFold 3

AlphaFold 3, released in May 2024 and co-developed by Google DeepMind and Isomorphic Labs, extended the capabilities of its predecessor by predicting the joint structure of complexes that include proteins, DNA, RNA, small molecules, ions, and modified residues.^[12][14] The model achieves at least a 50% improvement over existing methods for predicting protein interactions with other molecule types, and for some categories of interaction, accuracy has doubled compared to previous state-of-the-art approaches.^[12][15] The authors describe the result in the Nature paper: "we show that high-accuracy modelling across biomolecular space is possible within a single unified deep-learning framework."^[12] In tests reported by Isomorphic Labs, AlphaFold 3's protein-ligand docking accuracy on the PoseBusters benchmark reached 76.4%, about 1.8 times the 42% of RoseTTAFold All-Atom.^[16] For drug discovery, this matters because understanding how a drug molecule binds to its protein target is fundamental to rational drug design, and predicted binding sites accelerate both target assessment and the design of molecules that bind effectively. Pharmaceutical companies including Novartis and Eli Lilly are collaborating with Isomorphic to apply these capabilities to real-world drug design programs.^[17][18]

IsoDDE

In February 2026, Isomorphic Labs published a technical report describing the Isomorphic Labs Drug Design Engine (IsoDDE), which goes beyond structure prediction to model binding affinity, selectivity, antibody-antigen interfaces, and ligandable pocket identification.^[19][17] On a protein-ligand structure prediction generalization benchmark, IsoDDE roughly doubled the performance of AlphaFold 3: on the hardest cases, those with less than 20% similarity to training data, it reached 50% accuracy where AlphaFold 3 reached 23.3%.^[19][17] On a held-out set of 334 antibody-antigen complexes, IsoDDE reached 39% accuracy in the high-fidelity regime (DockQ greater than 0.8), which the report describes as 2.3 times the accuracy of AlphaFold 3 and 19.8 times that of Boltz-2, and the company reports that it outperforms both Boltz-2 and physics-based methods at predicting binding affinity.^[19][20] Mohammed AlQuraishi, a computational biologist at Columbia University, told Scientific American: "It's a major advance, on the scale of an AlphaFold4."^[20] The full architecture and weights have not been published.

Limitations

AlphaFold's database has made structural information available for virtually every known protein, opening up previously intractable targets for drug development, but there are important limitations.^[12] AlphaFold 2 predicts static structures and cannot fully capture the dynamic conformational changes that proteins undergo in biological environments, so the predicted structure may not represent the form that binds a drug molecule. AlphaFold also does not directly predict binding affinities or druggability, meaning that additional computational and experimental work is still needed to translate structural predictions into drug candidates.^[15]

The progression from AlphaFold 2 through AlphaFold 3 to IsoDDE shows a clear trajectory: from predicting what proteins look like, to predicting how molecules interact with proteins, to modeling the properties that decide whether a molecule can become a drug.

Molecule generation and de novo drug design

Once a target has been identified and validated, the next challenge is to find or create molecules that modulate it. Traditional high-throughput screening involves testing millions of compounds from physical libraries, a process that is expensive and limited to existing chemical space. AI-driven de novo drug design instead generates entirely new molecules from scratch, exploring a vastly larger chemical space.

Generative chemistry

Several generative model architectures have been adapted for molecular design. These include variational autoencoders (VAEs), generative adversarial networks, normalizing flows, and diffusion models. These models learn the statistical distribution of known drug-like molecules and can then sample from that distribution to generate novel compounds with specified properties.^[21]

A typical generative chemistry workflow involves encoding molecular structures as SMILES strings or molecular graphs, training a generative model on datasets of known bioactive compounds, and then sampling new molecules while optimizing for properties such as binding affinity, selectivity, solubility, and synthetic accessibility. Active learning cycles can iteratively refine the generated molecules using feedback from docking simulations, QSAR models, or even experimental assays.

Insilico Medicine's Chemistry42 platform uses generative models to design novel small molecules for a given target. Using this platform, Insilico designed rentosertib (ISM001-055) in a fraction of the time that traditional medicinal chemistry would require. The compound went from target identification to preclinical candidate nomination in approximately 18 months, compared to the industry average of 4 to 5 years.^[11]

Exscientia, a UK-based company that combined with Recursion Pharmaceuticals in November 2024, developed the Centaur AI platform, which combines generative design with automated experimentation. Exscientia reported that its platform reduced the drug design timeline from the industry average of 4.5 years to just 12 to 15 months while decreasing capital costs by up to 80%.^[22] Its automated precision chemistry used active learning and Bayesian optimization to iteratively design, synthesize, and test compounds, reducing the number of design-make-test cycles required.^[23]

Molecular diffusion models

One of the most active areas of research involves diffusion models adapted from the image generation domain.^[24] Diffusion models for molecular generation learn to reverse a noise-addition process: during training, the model learns to progressively denoise a random distribution of atoms back into valid molecular structures. By conditioning this process on target properties (binding affinity to a specific protein pocket, desired solubility, low toxicity), the model can generate novel molecules tailored to specific requirements.^[25]

Several notable generative chemistry systems emerged in 2025 and 2026:

A key challenge for generative molecular models remains ensuring chemical synthesizability. A model may design a molecule with ideal predicted properties, but if the molecule cannot be practically synthesized in a laboratory, it has limited value. Current research is focused on incorporating synthetic feasibility constraints directly into the generative process.^[24] Researchers have also identified unified molecular foundation models as a frontier: such models could design small molecules, peptides, and hybrid therapeutics from a single architecture rather than requiring separate specialized models for each molecular class.

Multi-objective optimization

A key challenge in drug design is that molecules must satisfy many conflicting requirements simultaneously. They need high potency against the target, selectivity over related proteins, favorable pharmacokinetic properties, low toxicity, and practical synthetic routes. AI methods are well suited to multi-objective optimization, using techniques such as Pareto optimization and reinforcement learning to navigate these trade-offs and identify molecules that represent the best compromises across all criteria.

Virtual screening

Virtual screening uses computational methods to evaluate large libraries of compounds and identify those most likely to bind to a target protein. There are two main approaches: structure-based virtual screening, which uses the three-dimensional structure of the target, and ligand-based virtual screening, which uses information about known active molecules.

Molecular docking

Molecular docking is the traditional workhorse of structure-based virtual screening. Docking programs predict how a small molecule fits into the binding site of a protein and estimate the strength of the interaction through a scoring function. However, traditional scoring functions often fail to accurately model the full complexity of protein-ligand interactions, leading to high false-positive rates.

AI-enhanced docking methods address these limitations. Machine learning scoring functions trained on large datasets of experimental binding data can better predict actual binding affinities. For instance, the Deep Docking (DD) platform enables up to 100-fold acceleration of structure-based virtual screening by docking only a subset of a chemical library and iteratively predicting the scores for remaining compounds using a ligand-based model. RosettaVS, published in Nature Communications, combines molecular dynamics with AI to model receptor flexibility and outperforms traditional methods on multiple benchmarks.^[26]

Ultra-large library screening

Recent advances have made it possible to screen virtual libraries containing billions of compounds. Atomwise's AtomNet platform uses deep learning and convolutional neural networks to analyze protein structures and predict drug-target interactions across a proprietary library of more than three trillion synthesizable compounds. By predicting docking scores with machine learning rather than performing full docking simulations, these platforms can evaluate chemical spaces that would be computationally prohibitive with traditional methods.

AI-guided antimicrobial peptide discovery

Antimicrobial peptide research applies a different search space from conventional small-molecule screening. Models inspect or generate amino-acid sequences, then researchers synthesize selected peptides and test their activity, toxicity, mechanism, and behavior in animal infection models. Computational predictions narrow the search, but a predicted or preclinical candidate is not an approved antibiotic.^[27][28][29]

Research led by Cesar de la Fuente at the University of Pennsylvania has progressed from sequence mining to constrained peptide optimization. The three published studies used related computational models and separate experimental pipelines.^[27][28][29]

The ApexGO results concern peptide optimization rather than an unrestricted search for any possible molecule. The model used the earlier APEX predictor as its objective function, proposed changes to known templates, and retained constraints on similarity to those templates. Its reported hit rates came from laboratory tests of synthesized peptides, while the infection results came from mice. They do not establish safety or efficacy in humans.^[29]

OpenAI published a customer story about the same laboratory on September 9, 2026, followed by a companion video titled "Discovering new antibiotics with ChatGPT" on September 10.^[30][31] In the written story, lab members described ChatGPT as supporting idea generation and knowledge sharing, and OpenAI Codex as helping researchers write scripts for analysis and consistent plotting.^[30] The story also quotes de la Fuente saying that the lab can discover molecules in hours rather than five or six years, but it does not define that comparison or provide a controlled measurement.^[30]

The OpenAI materials do not connect ChatGPT or Codex to a particular validated peptide or publish a before-and-after evaluation of scientific output. The peer-reviewed candidate counts and assay results above are attributed to APEX, APEX 1.1, ApexGO, and the accompanying wet-lab experiments. Taken together, the sources support treating ChatGPT and Codex as reported workflow tools in this case, separate from the scientific models whose outputs were experimentally tested.^{[30][27][28][29]}

ADMET prediction

ADMET stands for Absorption, Distribution, Metabolism, Excretion, and Toxicity. These pharmacokinetic and safety properties determine whether a drug candidate can reach its target in the body at an effective concentration without causing unacceptable side effects. Poor drug-like properties are responsible for roughly 10% to 15% of clinical trial failures and have historically been a major cause of late-stage attrition, which makes early ADMET evaluation critical.^[7]

Machine learning models for ADMET prediction have become increasingly sophisticated. Common approaches include:

Modern ADMET prediction frameworks increasingly use graph neural networks that operate directly on molecular graphs, capturing structural features that are relevant to pharmacokinetic behavior. Multitask learning, where a single model predicts multiple ADMET endpoints simultaneously, has shown improvements over single-task models because ADMET properties are often correlated. AutoML methods that automatically search for optimal model architectures and hyperparameters have further improved the reproducibility and performance of ADMET predictions.^[32] Frameworks such as ADMET-AI perform rapid, large-scale ADMET profiling, processing over a million molecules in a few hours; its developers report that it outperforms existing tools in both speed and accuracy and achieves top scores on benchmark datasets.^[33]

Despite these advances, challenges remain. Training data for ADMET models is often limited, noisy, and biased toward certain chemical series. Predictions may not generalize well to novel chemical scaffolds, and the gap between in vitro measurements and in vivo outcomes remains significant.^[32]

Clinical trial optimization

AI is increasingly being applied beyond drug design to the clinical trial process itself. Clinical trials account for approximately 60% of total drug development costs and are the stage most prone to delays and failures.

Patient recruitment

Nearly 80% of clinical trials experience delays, with patient recruitment cited as one of the most common challenges. AI platforms can identify protocol-eligible patients up to three times faster by processing electronic health records, genetic profiles, and demographic information with approximately 93% accuracy. Natural language processing models parse unstructured medical records to match patients to eligibility criteria, while predictive models identify patients most likely to benefit from treatment and to remain compliant throughout the trial.

Adaptive trial design

Adaptive clinical trials use accumulating data to modify aspects of the study without undermining its validity. Machine learning algorithms analyze interim results using reinforcement learning, decision trees, and neural networks to inform real-time protocol updates. These updates can include dosage adjustments, treatment arm modifications, and patient reallocation. AI-driven adaptive designs can reduce sample sizes, shorten trial durations, and increase the probability of detecting true treatment effects.

Digital twins and synthetic control arms

Digital twins, virtual representations of individual patients built from their clinical and molecular data, allow researchers to simulate patient-specific responses to treatment. Synthetic control arms use historical patient data modeled by AI to serve as comparators, potentially reducing the number of patients needed in control groups. Both approaches are being explored as ways to make trials more efficient and ethical, particularly for rare diseases where patient populations are small.

Which companies lead AI drug discovery?

The landscape includes AI-native biotech companies, established pharmaceutical firms building internal AI capabilities, and technology companies extending their AI platforms into drug discovery.^[34]

Isomorphic Labs

Isomorphic Labs was founded in 2021 as a spin-off from Google DeepMind, led by DeepMind founder Demis Hassabis. The company's mission is to apply AlphaFold and related AI technologies to drug discovery at scale, with an emphasis on structure prediction and molecular design. In January 2024, Isomorphic announced partnerships with Eli Lilly and Novartis worth up to nearly $3 billion in combined milestone payments.^[18][35] The Novartis partnership was expanded in February 2025 to include up to three additional research programs.^[36]

In March 2025, Isomorphic raised $600 million in its first external financing round, led by Thrive Capital with participation from GV and follow-on capital from existing investor Alphabet.^[37] The company stated the funding would "accelerate Isomorphic's frontier AI research and rapidly advance the company's next-generation AI drug design engine," with first programs heading toward the clinic.^[37] In July 2025, Fortune reported that Isomorphic was staffing up for its first human clinical trials, with lead candidates in oncology and immune-mediated disorders, and that its first molecules could enter Phase I by late 2026.^[38] In February 2026, Isomorphic unveiled IsoDDE, described in the AlphaFold section above.^[17][19]

Insilico Medicine

Insilico Medicine was founded in 2014 by Alex Zhavoronkov and is headquartered in Hong Kong. The company's end-to-end Pharma.AI platform includes PandaOmics for target identification, Chemistry42 for generative drug design, and InClinico for clinical trial outcome prediction. Since 2021, Insilico has nominated over 20 preclinical candidates across a portfolio of more than 30 assets and received IND clearance for 10 molecules.^[11] The company raised $110 million in a Series E round led by Value Partners in March 2025, and listed on the Hong Kong Stock Exchange on December 30, 2025, raising about HK$2.28 billion.^[39][40]

Insilico's lead program, rentosertib (originally ISM001-055), is a TNIK inhibitor for idiopathic pulmonary fibrosis and the first drug for which both the biological target and the molecule were discovered using generative AI to enter Phase II clinical trials. Phase IIa results from the GENESIS-IPF trial were published in Nature Medicine on June 3, 2025: in the randomized, double-blind, placebo-controlled study of 71 patients across 22 sites in China, patients receiving 60 mg once daily showed a mean improvement in forced vital capacity (FVC) of +98.4 mL over 12 weeks, compared to a mean decline of -20.3 mL in the placebo group.^[4] The most common drug-related adverse events were diarrhea (14.8%) and abnormal liver function (14.8%). Zhavoronkov said the results "not only suggest that Rentosertib has a manageable safety and tolerability profile, but also warrants further investigation in larger-scale clinical trials of longer duration, demonstrating the transformative potential of AI in drug discovery and development."^[4] A separate Phase IIa trial in the United States is also ongoing.

Insilico reports that across 2021 to 2024 it advanced 22 candidates from project initiation to preclinical candidate nomination in 12 to 18 months each, synthesizing only about 60 to 200 molecules per program, versus the 2.5 to 4 years and thousands of compounds typical of traditional discovery.^[4] Beyond rentosertib, its pipeline includes ISM3091, a USP1 inhibitor for solid tumors licensed to Exelixis, and ISM5939, an oral ENPP1 inhibitor that received FDA IND clearance in November 2024 for a first-in-human Phase 1a/b study in solid tumors.^[41]

Recursion Pharmaceuticals

Recursion Pharmaceuticals was founded in 2013 in Salt Lake City, Utah. The company operates highly automated laboratories that run up to 2.2 million experiments per week, imaging cells under millions of experimental conditions and feeding the resulting phenomic data into the Recursion Operating System (Recursion OS), which uses AI to identify patterns that suggest therapeutic opportunities. Over the past decade, Recursion has accumulated over 50 petabytes of proprietary biological and chemical data spanning phenomics, transcriptomics, proteomics, and ADME measurements.^[42]

On November 20, 2024, Recursion completed its combination with Exscientia, an all-stock deal valued at about $688 million when it was announced in August 2024, combining Recursion's phenomic screening with Exscientia's automated precision chemistry into an end-to-end AI drug discovery platform.^[23][43] In May 2025, Recursion narrowed its pipeline, discontinuing REC-994 (cerebral cavernous malformation) and REC-2282 (NF2) and seeking an out-licensing partner for REC-3964, a Clostridioides difficile Toxin B inhibitor.^[44]

Key clinical programs include:

• REC-4881, a MEK1/2 inhibitor for familial adenomatous polyposis (FAP), which showed a median 43% reduction in polyp burden at week 13 in the Phase 1b/2 TUPELO trial; as of mid-2026 the trial was enrolling an expanded cohort, with an update on a registrational path expected in the second half of 2026.^[45][46]
• REC-617 (formerly GTAEXS617, from Exscientia), an oral CDK7 inhibitor for advanced solid tumors in Phase 1, with combination data expected in the first half of 2027.^[47][46]
• REC-1245, an RBM39 degrader for biomarker-enriched solid tumors and lymphoma. Recursion says it went from target identification to IND-enabling studies in under 18 months with about 200 compounds synthesized; the first patient in its Phase 1/2 DAHLIA study was dosed in December 2024.^[48]

In partnership with NVIDIA, Recursion built BioHive-2, described in the infrastructure section below.^[49]

Exscientia (now part of Recursion)

Oxford-based Exscientia pioneered the use of AI to design small-molecule drug candidates. In January 2020, DSP-1181, a candidate for obsessive-compulsive disorder developed with Sumitomo Dainippon Pharma, entered a Phase I study in Japan, making it the first AI-designed molecule to enter human clinical trials.^[50] Before its combination with Recursion, six molecules designed using Exscientia's AI entered clinical trials.^[22] Notable programs included EXS21546, an A2A receptor antagonist for immuno-oncology that was discovered in eight months and was the first AI-designed oncology immunotherapy molecule to enter human trials, but was dropped in Exscientia's October 2023 pipeline reprioritization; DSP-0038, a dual 5-HT1a/5-HT2a modulator for Alzheimer's disease psychosis developed with Sumitomo; and GTAEXS617, the CDK7 inhibitor now developed by Recursion as REC-617.^[22][51][46] Exscientia's technology and pipeline are now integrated into the Recursion platform.^[23]

BenevolentAI

London-based BenevolentAI focuses on using AI for target identification and drug repurposing. In February 2020, early in the COVID-19 pandemic, its researchers used the company's knowledge graph to identify baricitinib, an existing JAK inhibitor, as a potential COVID-19 treatment.^[52] The prediction was subsequently tested in clinical trials: the FDA issued an emergency use authorization for baricitinib (in combination with remdesivir) for hospitalized COVID-19 patients in November 2020 and granted full approval for hospitalized adults in May 2022.^[53] BenevolentAI's own drug candidate BEN-2293, a topical treatment for atopic dermatitis, met its safety endpoint in a Phase IIa trial in April 2023 but missed its efficacy endpoints in the overall study population.^[54]

After financial challenges, BenevolentAI announced a strategic overhaul in December 2024 and a plan to explore delisting. Shareholders approved a merger into Osaka Holdings, a vehicle created for the transaction, in March 2025, and the company delisted from Euronext Amsterdam on March 13, 2025.^[55][56] Its trajectory illustrates the commercial challenges facing AI drug discovery companies even when the underlying technology produces validated results.

Relay Therapeutics

Relay Therapeutics is a clinical-stage precision medicine company founded in Cambridge, Massachusetts. Its Dynamo platform integrates molecular dynamics simulations, cryo-electron microscopy, and machine learning to design drugs that target specific protein conformations.^[57] This approach enables the creation of mutant-selective inhibitors that target only the pathological form of a protein while minimizing off-target effects.

Relay's lead asset, RLY-2608, is a pan-mutant PI3K-alpha inhibitor. In a Phase 1b study in combination with fulvestrant, presented in 2025, patients with PI3K-alpha-mutated HR+/HER2- metastatic breast cancer had a 10.3-month median progression-free survival and a 39% overall response rate; the Phase 3 ReDiscover-2 trial began in the second quarter of 2025.^[58][59] In December 2024 the company licensed RLY-4008 (lirafugratinib) to Elevar Therapeutics in a deal worth up to $500 million.^[60] Relay also has a collaboration with Pfizer to explore combination therapies.

Schrodinger

Schrodinger, Inc. was founded in 1990 in New York City and is a pioneer in physics-based computational chemistry for drug discovery.^[61] Rather than relying solely on data-driven machine learning, its platform uses first-principles physics simulations (molecular dynamics, free energy calculations, and quantum mechanics) to predict how molecules interact, and integrates predictive modeling, data analytics, and collaboration tools for rapid exploration of chemical space.^[62] While the company has sometimes resisted the "AI" label, its platform increasingly incorporates machine learning, which lets it evaluate billions of virtual compounds, and it has been expanding into physics-guided generative workflows for molecular design.^[63]

The most advanced clinical molecule to come out of Schrodinger's platform is zasocitinib (TAK-279), a highly selective allosteric TYK2 inhibitor originally developed by Nimbus Therapeutics using Schrodinger's computational platform. Takeda acquired the program from Nimbus in December 2022 for $4 billion upfront plus up to $2 billion in milestones, and Schrodinger, a Nimbus shareholder, received a $111.3 million distribution from the sale.^[64] Zasocitinib demonstrates more than one-million-fold selectivity for TYK2 over JAK1, JAK2, and JAK3.^[65] In December 2025, Takeda reported positive Phase 3 topline results from its LATITUDE program in plaque psoriasis, with more than half of patients achieving clear or almost clear skin (PASI 90) and about 30% achieving completely clear skin (PASI 100) at week 16; the studies met both co-primary endpoints and all 44 ranked secondary endpoints.^[65][66] The FDA accepted Takeda's new drug application for priority review in September 2026, with a decision expected in the first quarter of 2027.^[67] Zasocitinib is also being evaluated in Phase 3 studies for psoriatic arthritis and Phase 2 studies for inflammatory bowel disease.

Pharmaceutical AI computing infrastructure

Several pharmaceutical companies have invested in dedicated AI computing infrastructure rather than relying solely on cloud resources, a strategic bet that computational scale will be a competitive advantage in drug discovery.

Recursion BioHive-2

In partnership with NVIDIA, Recursion built BioHive-2, an NVIDIA DGX SuperPOD comprising 63 DGX H100 systems with 504 NVIDIA H100 Tensor Core GPUs that delivers about 2 exaflops of AI performance. Completed in May 2024, it ranked No. 35 on the TOP500 list and was at the time the fastest supercomputer wholly owned by a pharmaceutical company.^[49]

Eli Lilly's LillyPod

In late February 2026, Eli Lilly inaugurated LillyPod, described as the most powerful AI factory wholly owned and operated by a pharmaceutical company and the world's first NVIDIA DGX SuperPOD built with DGX B300 systems.^[68][69] LillyPod is powered by 1,016 NVIDIA Blackwell Ultra GPUs, delivering more than 9,000 petaflops of AI performance; the system was assembled in about four months and inaugurated at a ceremony in Indianapolis.^[70]

LillyPod is designed to support drug discovery, genomics, clinical development, and manufacturing optimization. The system creates what Lilly calls a "computational dry lab at massive scale," where scientists can simulate and evaluate billions of molecular hypotheses in parallel before committing to physical experiments. Lilly employees can also use LillyPod to build chatbots, agentic workflows, and research lab agents using the company's internal AI platforms.^[68]

Do AI-designed drugs actually have higher success rates?

The number of AI-designed drug candidates entering clinical trials has grown rapidly, from 3 programs in 2016 to 67 in 2023 and more than 173 as of early 2026.^[2] The first systematic analysis of this pipeline, by Jayatunga and colleagues in Drug Discovery Today (2024), examined the AI-discovered molecules that had completed Phase I as of December 2023 and found that 21 of 24 succeeded, an 80% to 90% success rate that is "substantially higher than historic industry averages" of roughly 40% to 65%.^[1] The authors concluded this "suggests that AI is highly capable of designing or identifying molecules with drug-like properties."^[1]

The same analysis introduced an important caveat that drives the ongoing debate: in Phase II, AI-discovered drugs succeeded at only about 40% (4 of 10), in line with the historic industry range, although on a small sample.^[1] In other words, the strongest evidence to date shows AI improving the odds of clearing the earliest safety stage rather than guaranteeing efficacy in patients. As of October 2026, no AI-discovered drug has received FDA approval; the computationally designed zasocitinib is under FDA priority review with a decision expected in early 2027, and research aggregators estimate that over 200 AI-originated programs are in some stage of clinical development, with 15 to 20 expected to enter pivotal (Phase III) trials in 2026.^[67][2]

The table below highlights notable AI-discovered or AI-designed drugs that have reached clinical trials.^[71]

Challenges and limitations

Data quality and availability

AI models are only as good as the data they are trained on. In drug discovery, high-quality labeled data is scarce: despite the vast scale of existing chemical libraries, rigorously curated datasets with robust biological, pharmacological, and clinical annotations remain rare. Public databases such as ChEMBL, PubChem, and BindingDB contain millions of data points, but these data are often noisy, inconsistent, or biased toward certain target classes, and results can vary across laboratories and experimental conditions. A survey of technology executives found that 68% cite poor data quality and governance as the primary reason AI initiatives fail, and data preparation alone consumes an estimated 80% of an AI project's time. Much of the most valuable data (clinical outcomes, proprietary compound activity) is held by pharmaceutical companies and is rarely shared because of competitive concerns, limiting the ability to train broadly generalizable models.

Biological complexity and the validation gap

Biology is far more complex than current AI models can fully capture. Diseases involve intricate networks of interacting proteins, genes, and environmental factors. A molecule that looks promising in silico may fail in living systems due to off-target effects, unexpected metabolic pathways, or immune responses that no model predicted. Predictions made by AI models (virtual screening hits, predicted binding affinities, generated molecules) must ultimately be validated through wet-lab experiments and clinical trials, and bridging that gap requires tight integration between computational and experimental teams along with realistic expectations about what AI can and cannot predict. The gap between computational predictions and clinical reality remains the fundamental challenge in AI-driven drug discovery, and it is the most plausible explanation for why AI's Phase II success rates have so far matched, rather than exceeded, traditional approaches.^[1]

Regulatory landscape

On January 7, 2025, the FDA published its first comprehensive draft guidance on the use of AI in drug development, titled "Considerations for the Use of Artificial Intelligence to Support Regulatory Decision-Making for Drug and Biological Products." The guidance introduces a seven-step credibility assessment framework based on "context of use," requiring lifecycle maintenance plans and transparency about model architectures and training data.^[72] Different levels of documentation are required depending on risk level, which the framework derives from model influence and decision consequence. The FDA's Center for Drug Evaluation and Research (CDER) has also been engaging directly with AI drug discovery companies.^[2]

The regulatory framework remains a work in progress. The FDA has indicated that AI-discovered drugs will be evaluated by the same safety and efficacy standards, and through the same approval pathway, as traditionally discovered drugs, but the use of AI in regulatory submissions raises open questions: how much detail about the AI methods must be disclosed, how AI-derived evidence will be weighted, how regulators will handle the rapid pace of model evolution, and how to ensure interpretability, reproducibility, and accountability. More than 25% of warning letters issued by the FDA since 2019 have cited data accuracy issues, underscoring the importance of data governance in AI applications.

Interpretability, reproducibility, and transparency

Many AI models, particularly deep neural networks, function as "black boxes" that provide predictions without transparent explanations of their reasoning. In a field where understanding the mechanism of action is critical for both scientific progress and regulatory approval, the lack of interpretability is a significant barrier. Efforts to develop explainable AI methods for drug discovery are ongoing but have not yet produced solutions that satisfy both computational performance and human interpretability requirements.

Many AI drug discovery publications also report impressive performance on benchmark datasets, but reproducibility in real-world drug discovery settings has been harder to demonstrate. The field has been criticized for an overreliance on retrospective analyses and in silico benchmarks rather than prospective experimental validation. The secrecy surrounding proprietary systems such as IsoDDE, whose full architecture and weights have not been published, has itself become a point of debate in the research community.^[20]

Commercial viability

The path to commercial viability for AI drug discovery companies has proven challenging. BenevolentAI's restructuring and delisting, despite producing a validated drug-repurposing prediction, illustrates the gap between scientific achievement and commercial sustainability. Drug discovery remains a high-risk, long-timeline business, and AI has not yet closed the gap between promising preclinical results and approved products generating revenue.^[56] Industry consolidation, such as the Recursion-Exscientia combination, reflects a push toward end-to-end platforms rather than narrow AI applications.^[23]

How big is the AI drug discovery market?

The AI in drug discovery market is growing, though estimates vary widely across research firms depending on scope and methodology:

Sources: ^[9][73][74].

According to Grand View Research, North America accounted for about 53% of revenue in 2025, the drug optimization and repurposing segment led by application with roughly 52% share, and the preclinical stage is the fastest-growing segment.^[9] Oncology is the most prominent therapeutic area using AI tools, followed by immunology, neurology, and infectious disease. Key drivers include the rising demand for cost-effective drug development, the increasing number of clinical trials, and growing chronic disease prevalence.^[9]

Major technology companies have made substantial investments in the space. Alphabet invested in Isomorphic Labs, which raised $600 million in March 2025.^[37] NVIDIA has partnered with Recursion Pharmaceuticals on the BioHive-2 supercomputer.^[49] AMD invested $20 million in Absci in January 2025 to support AI-driven biologics discovery. These investments reflect the growing conviction that AI will become a foundational technology in pharmaceutical R&D.

Current state and outlook

The field of AI in drug discovery is at an inflection point. Several trends define the landscape as of 2026:

Clinical validation milestones. Insilico Medicine's Nature Medicine publication of the rentosertib Phase IIa results (June 2025) was the first proof-of-concept clinical validation of a drug whose target and molecule were both found with generative AI.^[4] With rentosertib showing positive Phase IIa results in IPF and zasocitinib under FDA review, the field is beginning to show that computationally discovered drugs can succeed in the clinic, and the first approval of such a drug is widely anticipated in 2026 or 2027.^[67][2]

Industry consolidation. The Recursion-Exscientia combination signals a trend toward consolidation, as companies seek to build end-to-end platforms rather than focusing on narrow AI applications.^[23] This mirrors broader patterns in the pharmaceutical industry, where scale and integration are competitive advantages.

Pharma investment in AI. Eli Lilly's LillyPod and Isomorphic Labs' partnerships with Lilly and Novartis, worth up to nearly $3 billion, show that major pharmaceutical companies view AI as a strategic priority rather than an experimental tool.^[35][68]

Next-generation models. The progression from AlphaFold 2 to AlphaFold 3 to IsoDDE represents a rapid evolution in AI capabilities for drug discovery.^[20] Generative chemistry models, particularly diffusion-based approaches, are advancing quickly. Other trends to watch include foundation models for biology that integrate diverse data types into unified representations, federated learning approaches that allow pharmaceutical companies to train models on pooled data without sharing proprietary information, and autonomous AI agents that can design, execute, and interpret experiments with minimal human oversight.

Persistent challenges. Data quality, experimental validation, regulatory frameworks, and commercial viability remain significant obstacles. The field has learned that while AI can dramatically accelerate certain stages of discovery, drug development as a whole involves challenges (clinical trial logistics, manufacturing, regulatory navigation, commercial strategy) that AI alone cannot solve. The most successful companies integrate AI tightly with experimental science and clinical development rather than treating it as a standalone solution. No amount of algorithmic sophistication can fully substitute for experimental validation, and the ultimate test of AI-designed drugs will be their performance in the clinic.

See also

• AI in healthcare
• AI for science
• AlphaFold
• Protein structure prediction
• Computational chemistry
• Molecular dynamics

References

1. ^1 ^2 ^3 ^4 ^5 ^6Jayatunga, M. K. P. et al. "How successful are AI-discovered drugs in clinical trials? A first analysis and emerging lessons." Drug Discovery Today 29, no. 6 (2024): 104009. doi.org/...j.drudis.2024.104009
2. ^1 ^2 ^3 ^4 ^5Axis Intelligence. "AI Drug Discovery 2026: 173 Programs, FDA Framework & Market." 2026. axis-intelligence.com/...ry-2026-complete-analysis
3. ^1 ^2 ^3"Demis Hassabis & John Jumper awarded Nobel Prize in Chemistry." Google DeepMind, October 2024. deepmind.google/...warded-nobel-prize-in-chemistry
4. ^1 ^2 ^3 ^4 ^5 ^6Insilico Medicine. "Insilico Medicine Announces Nature Medicine Publication of Phase IIa Results Evaluating Rentosertib." PR Newswire, June 2025. prnewswire.com/...pioneering-ai-approach-302472070
5. ^DiMasi, J. A. et al. "Innovation in the pharmaceutical industry: New estimates of R&D costs." Journal of Health Economics 47 (2016): 20-33. doi.org/...j.jhealeco.2016.01.012
6. ^"New drugs cost US$2.6 billion to develop." *Nature Reviews Drug Discovery*, 2014. nature.com/...nrd4507
7. ^1 ^2Sun, D. et al. "Why 90% of clinical drug development fails and how to improve it?" Acta Pharmaceutica Sinica B 12, no. 7 (2022): 3049-3062.
8. ^1 ^2Biotechnology Innovation Organization, Informa Pharma Intelligence, and QLS Advisors. "Clinical Development Success Rates 2011-2020." February 2021. go.bio.org/...DevelopmentSuccessRates2011_2020.pdf
9. ^1 ^2 ^3 ^4Grand View Research. "Artificial Intelligence In Drug Discovery Market Report, 2033." grandviewresearch.com/...nce-drug-discovery-market
10. ^"Leading artificial intelligence-driven drug discovery platforms: 2025 landscape and global outlook." *ScienceDirect*, 2025. sciencedirect.com/...S0031699725075118
11. ^1 ^2 ^3Insilico Medicine. "First Generative AI Drug Begins Phase II Trials with Patients." insilico.com, 2023.
12. ^1 ^2 ^3 ^4 ^5 ^6Abramson, J. et al. "Accurate structure prediction of biomolecular interactions with AlphaFold 3." Nature 630 (2024): 493-500.
13. ^The Nobel Foundation. "The Nobel Prize in Chemistry 2024." nobelprize.org, October 2024.
14. ^"AlphaFold 3 ushers in a new era for biomedical research and drug discovery." EurekAlert, 2024. eurekalert.org/...1101499
15. ^1 ^2Callaway, E. "Major AlphaFold upgrade offers boost for drug discovery." Nature (2024).
16. ^"Rational drug design with AlphaFold 3." Isomorphic Labs. isomorphiclabs.com/...drug-design-with-alphafold-3
17. ^1 ^2 ^3 ^4Isomorphic Labs. "The Isomorphic Labs Drug Design Engine unlocks a new frontier beyond AlphaFold." isomorphiclabs.com, February 2026.
18. ^1 ^2CNBC. "Inside Isomorphic Labs, the secretive AI life sciences startup spun off from Google DeepMind." cnbc.com, April 2025.
19. ^1 ^2 ^3 ^4Isomorphic Labs. "Accurate Predictions of Novel Biomolecular Interactions" (IsoDDE technical report). isomorphiclabs.com, February 10, 2026.
20. ^1 ^2 ^3 ^4"An AlphaFold 4: scientists marvel at DeepMind drug spin-off's exclusive new AI." *Scientific American* / *Nature*, 2026. scientificamerican.com/...in-offs-exclusive-new-ai
21. ^Pun, F. W. et al. "A survey of generative AI for de novo drug design: new frontiers in molecule and protein generation." Briefings in Bioinformatics 25, no. 4 (2024).
22. ^1 ^2 ^3Exscientia / UKRI. "Exscientia: a clinical pipeline for AI-designed drug candidates." ukri.org.
23. ^1 ^2 ^3 ^4 ^5"Recursion and Exscientia have officially combined." Recursion Pharmaceuticals, 2025. ir.recursion.com/...eaders-ai-drug-discovery-space
24. ^1 ^2"Generative AI for the Design of Molecules: Advances and Challenges." *Journal of Chemical Information and Modeling*, 2025. pubs.acs.org/...acs.jcim.5c02234
25. ^"Diffusion Models in De Novo Drug Design." *Journal of Chemical Information and Modeling*, 2024. pubs.acs.org/...acs.jcim.4c01107
26. ^Nature Communications. "An artificial intelligence accelerated virtual screening platform for drug discovery." (2024).
27. ^1 ^2 ^3 ^4Wan, F., Torres, M. D. T., Peng, J., and de la Fuente-Nunez, C. "Deep-learning-enabled antibiotic discovery through molecular de-extinction." Nature Biomedical Engineering 8 (2024): 854-871.
28. ^1 ^2 ^3 ^4Torres, M. D. T., Wan, F., and de la Fuente-Nunez, C. "Deep learning reveals antibiotics in the archaeal proteome." Nature Microbiology 10 (2025): 2153-2167.
29. ^1 ^2 ^3 ^4 ^5Torres, M. D. T. et al. "A generative artificial intelligence approach for peptide antibiotic optimization." Nature Machine Intelligence 8 (2026): 841-856.
30. ^1 ^2 ^3 ^4OpenAI. "Accelerating antibiotic discovery with ChatGPT." September 9, 2026.
31. ^OpenAI. "Discovering new antibiotics with ChatGPT." YouTube video, September 10, 2026.
32. ^1 ^2"Leveraging machine learning models in evaluating ADMET properties for drug discovery and development." PMC (2025).
33. ^"Revolutionizing pharmacology: AI-powered approaches in ADMET prediction." *ScienceDirect*, 2025. sciencedirect.com/...S259009862500020X
34. ^"Top AI Drug Discovery Companies in 2026." PharmaNow. pharmanow.live/...ai-drug-discovery-companies-2026
35. ^1 ^2"The $3 Billion Bet: How Isomorphic Labs Is Rewriting Drug Discovery with Eli Lilly and Novartis." FinancialContent, 2026. markets.financialcontent.com/...lilly-and-novartis
36. ^Isomorphic Labs. "Isomorphic Labs announces Novartis collaboration expansion." February 18, 2025. isomorphiclabs.com/...rtis-collaboration-expansion
37. ^1 ^2 ^3Isomorphic Labs. "Isomorphic Labs announces $600m external investment round." isomorphiclabs.com, March 2025.
38. ^"Isomorphic Labs has grand ambitions to 'solve all diseases' with AI." Fortune, 2025. fortune.com/...-diseases-ai-now-first-human-trials
39. ^BioSpectrum Asia. "Insilico Medicine secures $110 M Series E financing to advance AI-driven drug discovery innovation." March 2025. biospectrumasia.com/...n-drug-discovery-innovation
40. ^Cooley. "Insilico Medicine announces HK$2.28 billion IPO." December 30, 2025. cooley.com/...dicine-announces-hk$2-28-billion-ipo
41. ^Exelixis. "Exelixis and Insilico Medicine Enter into Exclusive Global License Agreement for ISM3091." ir.exelixis.com.
42. ^Recursion Pharmaceuticals. "Pioneering AI Drug Discovery." recursion.com.
43. ^Scottish Financial Review. "Dundee spinout Exscientia bought for $688m in shares." August 9, 2024. scottishfinancialreview.com/...-for-688m-in-shares
44. ^Yahoo Finance. Report on Recursion discontinuing drug programs to streamline its pipeline, May 2025. finance.yahoo.com/...ogrammes-streamline-150100638
45. ^Recursion. "Positive Phase 1b/2 Results from Ongoing REC-4881 TUPELO Trial." ir.recursion.com, 2025.
46. ^1 ^2 ^3Recursion Pharmaceuticals. "Second quarter 2026 financial results and business update" (Exhibit 99.1, Form 8-K). SEC, August 2026. sec.gov/...exhibit991-q0226
47. ^Recursion. "Recursion Reports Interim Phase 1 Clinical Data for REC-617." ir.recursion.com, 2025.
48. ^"Recursion announces first patient dosed in Phase 1/2 clinical study of REC-1245." Recursion Pharmaceuticals. ir.recursion.com/...-dosed-phase-12-clinical-study
49. ^1 ^2 ^3NVIDIA Blog. "Drug Discovery, STAT! NVIDIA, Recursion Speed Pharma R&D With AI Supercomputer." blogs.nvidia.com, May 2024.
50. ^Sumitomo Dainippon Pharma. News release on the start of a Phase 1 study of DSP-1181 in Japan. January 30, 2020. sumitomo-pharma.com/...20200130
51. ^Fierce Biotech. "AI drug hunter Exscientia chops down rapidly emerging pipeline to focus on 2 main oncology assets." October 2023. fiercebiotech.com/...ipeline-focus-2-main-oncology
52. ^Richardson, P. et al. "Baricitinib as potential treatment for 2019-nCoV acute respiratory disease." The Lancet 395, no. 10223 (2020): e30-e31. doi.org/...S0140-6736(20)30304-4
53. ^Eli Lilly and Company. Announcement of FDA approval of baricitinib (Olumiant) for certain hospitalized adults with COVID-19. May 2022. investor.lilly.com/...pdf
54. ^BenevolentAI. Announcement of top-line Phase IIa results for BEN-2293 in atopic dermatitis. Business Wire, April 4, 2023. businesswire.com/...en
55. ^BenevolentAI. "Result of Extraordinary General Meeting and delisting from Euronext Amsterdam." Business Wire via DutchNews, March 2025. dutchnews.nl/...-delisting-from-euronext-amsterdam
56. ^1 ^2"The AI drug breakthrough is taking a long time to arrive." Fortune, 2025. fortune.com/...e-to-do-with-the-technologys-limits
57. ^Relay Therapeutics. "Dynamo Platform." relaytx.com.
58. ^AInvest. "Relay Therapeutics (RLAY): Is H2 2025 the Inflection Point for AI-Driven Drug Discovery?" ainvest.com, 2025.
59. ^Relay Therapeutics. "Relay Therapeutics Reports Second Quarter 2025 Financial Results." 2025. ir.relaytx.com/...d-quarter-2025-financial-results
60. ^Relay Therapeutics and Elevar Therapeutics. "Exclusive global licensing agreement for lirafugratinib." BioSpace, December 3, 2024. biospace.com/...iocarcinoma-and-other-solid-tumors
61. ^Schrodinger, Inc. "About us." schrodinger.com.
62. ^"Schrodinger Computational Platform for Molecular Discovery." Schrodinger. schrodinger.com/platform
63. ^"NVIDIA Partners with Schrodinger to Accelerate Drug Discovery." NVIDIA Newsroom. nvidianews.nvidia.com/...-drug-discovery-worldwide
64. ^Schrodinger. "Schrodinger Receives $111.3 Million Distribution from Sale of Nimbus's TYK2 Inhibitor to Takeda." February 14, 2023. ir.schrodinger.com/...default
65. ^1 ^2Takeda. "Takeda's Zasocitinib Landmark Phase 3 Plaque Psoriasis Data." takeda.com, December 2025.
66. ^AJMC. "Zasocitinib Hits Phase 3 End Points in Plaque Psoriasis." ajmc.com, 2025.
67. ^1 ^2 ^3Takeda. Announcement of FDA acceptance of the zasocitinib new drug application for priority review in plaque psoriasis. September 2026. takeda.com/...riority-review-zasocitinib-psoriasis
68. ^1 ^2 ^3"Lilly Launches LillyPod NVIDIA DGX SuperPOD for Genomics and Drug Discovery AI." AIwire/HPCwire, 2026. hpcwire.com/...-for-genomics-and-drug-discovery-ai
69. ^"Lilly partners with NVIDIA to build the industry's most powerful AI supercomputer." Eli Lilly Investor Relations. investor.lilly.com/...d-industrys-most-powerful-ai
70. ^"Lilly Deploys World's Largest AI Factory for Drug Discovery Using NVIDIA Blackwell." NVIDIA Blog. blogs.nvidia.com/...-nvidia-blackwell-dgx-superpod
71. ^"AI Drugs Are Already in Clinical Trials." HumAI Blog. humai.blog/...the-most-important-test-for-medicine
72. ^FDA. "Considerations for the Use of Artificial Intelligence to Support Regulatory Decision-Making for Drug and Biological Products" (draft guidance). fda.gov, January 2025.
73. ^Precedence Research. "Artificial Intelligence (AI) in Drug Discovery Market Size." precedenceresearch.com/...in-drug-discovery-market
74. ^MarketsandMarkets. "AI in Drug Discovery Market" press release, July 2026 (via Webull News). webull.com/...15271438694376448