AI drug discovery: how frontier AI is transforming pharmaceutical research

01How AI accelerates drug discovery

Artificial intelligence (AI) and its subfields have been used in applications throughout industry and academia. Machine learning has been used for various scientific and commercial purposes, including language translation, image recognition, decision-making, credit scoring, and e-commerce. Since the 2020s, massive advancements have been made in the field of generative artificial intelligence (GenAI), which generates text, images, music, videos, and other forms of data.

Artificial intelligence has been applied to drug discovery since the 2000s, but the field transformed after 2020 when deep learning models became capable of predicting protein structures, generating novel molecular structures, and screening billions of compounds in silico. The traditional drug discovery pipeline takes 10-15 years and costs over $2 billion per approved drug. AI aims to compress this by automating the most time-consuming stages: target identification, lead compound generation, and optimization.

The acceleration comes from three capabilities. First, predictive modeling — AI can predict how a protein will fold and what molecules will bind to it, eliminating months of laboratory trial and error. Second, generative chemistry — models can design entirely novel molecular structures optimized for specific targets, expanding the chemical space far beyond what human chemists can explore. Third, virtual screening — AI can screen billions of compounds against a target in hours rather than the years required for high-throughput physical screening.

02From molecule screening to clinical trials

The pharmaceutical industry discovers, develops, produces, and markets pharmaceutical goods such as medications. The pipeline from molecule to medicine is long and failure-prone: approximately 90 percent of drugs that enter Phase I clinical trials never reach the market. AI is being deployed across every stage to improve the odds.

In target identification, AI analyzes genomic and proteomic data to identify disease-related proteins that drugs could target. In lead generation, generative models propose novel compounds with desired properties. In preclinical testing, AI models predict toxicity and pharmacokinetics, reducing the number of animal studies needed. In clinical trials, AI helps with patient stratification, trial design optimization, and real-time monitoring of adverse events.

03AlphaFold and protein structure prediction

AlphaFold is an artificial intelligence (AI) program developed by DeepMind, a subsidiary of Alphabet, which performs predictions of protein structure. It is designed using deep learning techniques.

AlphaFold, released by DeepMind in 2020, solved a problem that had stymied biology for 50 years: predicting a protein's three-dimensional structure from its amino acid sequence. The system uses deep learning to predict the distance and angle between every pair of amino acids, producing structural models with near-experimental accuracy. AlphaFold 2 achieved a median Global Distance Test score of 92.4 across all targets in the Critical Assessment of Protein Structure Prediction competition, a level previously considered impossible for computational methods.

The impact on drug discovery is direct. Knowing a protein's structure allows researchers to design molecules that fit into its active site, the way a key fits a lock. Before AlphaFold, determining a protein structure required X-ray crystallography or cryo-electron microscopy, processes that take months and cost hundreds of thousands of dollars per protein. AlphaFold's database now contains over 200 million predicted protein structures, covering nearly every known protein across all organisms. This represents the most significant expansion of structural biology knowledge in history.

Isomorphic Labs, a DeepMind spin-off, is using AlphaFold technology to design drugs entirely through computational methods. The company has partnerships with Novartis and Eli Lilly worth up to $3 billion in milestone payments, signaling that major pharmaceutical companies are betting on AI-designed drugs as the future of the industry.

04The economic case for AI in pharma

The pharmaceutical industry is a medical industry that discovers, develops, produces, and markets pharmaceutical goods such as medications. Medications are then administered to patients for curing or preventing disease or for alleviating symptoms of illness or injury.

The economic pressure on pharmaceutical companies is intensifying. The average cost of bringing a new drug to market exceeds $2 billion, and the failure rate means that each successful drug must also pay for nine failed ones. Patent cliffs — the expiration of patents on blockbuster drugs — are shrinking R&D budgets at major pharma companies. AI offers a path to reduce costs at every stage: cheaper target identification, faster lead optimization, better preclinical predictions, and more efficient clinical trials.

The market for AI in drug discovery is projected to grow from approximately $1.5 billion in 2023 to over $10 billion by 2030. Major pharmaceutical companies — Pfizer, Roche, Novartis, Sanofi, AstraZeneca — have all established AI research partnerships or internal AI divisions. The strategic logic is clear: a drug that reaches market one year earlier generates hundreds of millions in additional revenue before patent expiration.

05Which companies are leading AI drug discovery

The AI drug discovery landscape includes both dedicated AI companies and traditional pharma firms adopting AI. Among the pure-play AI companies, Insilico Medicine has brought an AI-designed drug for idiopathic pulmonary fibrosis into Phase II clinical trials, the first AI-discovered drug to reach this stage. Recursion Pharmaceuticals uses automated microscopy and machine learning to screen thousands of compounds against disease models, and went public in 2021 at a valuation exceeding $3 billion.

Exscientia, based in Oxford, was the first company to put an AI-designed drug into clinical trials in 2020. Isomorphic Labs, spun out of DeepMind, is building drug discovery on AlphaFold foundations. XtalPi combines quantum physics and AI for drug development and has raised over $700 million. Schrödinger offers physics-based computational drug discovery platforms used by most major pharma companies.

Among traditional pharma, Novartis, Eli Lilly, Bayer, and Sanofi have all signed multi-billion-dollar AI partnerships. The industry is bifurcating: companies that build AI capabilities internally versus those that partner with AI specialists. The latter is faster but risks dependency on external platforms for a core competitive capability.

06Regulatory pathways for AI-discovered drugs

The FDA has not established specific regulatory frameworks for AI-discovered drugs, treating them under existing drug approval pathways. The key question is whether the origin of a drug molecule — human-designed versus AI-generated — matters for safety and efficacy assessment. The FDA's position is that it evaluates the drug, not the process that created it. However, AI raises new questions about data integrity, reproducibility, and the transparency of the design process.

The FDA has approved AI-assisted medical devices and diagnostic tools, and several AI-designed drugs are in the approval pipeline. The regulatory challenge is not the AI itself but the novelty of the molecules it produces. AI can generate molecular structures that have never been synthesized before, meaning there is no historical safety data to reference. Regulators are developing frameworks for evaluating drugs where the mechanism of discovery is fundamentally different from traditional chemistry.

07What patients should expect from AI-driven medicine

For patients, the most immediate impact of AI drug discovery will be faster access to new treatments, particularly for diseases that have been commercially neglected because the patient population is too small to justify traditional R&D costs. AI lowers the cost of discovery, making it economically viable to develop drugs for rare diseases that affect fewer than 200,000 people in the United States.

AI is also enabling personalized medicine — drugs designed for specific genetic profiles rather than broad populations. By analyzing patient genomic data alongside disease biology, AI can identify which patients will respond to which treatments, reducing the trial-and-error approach that characterizes much of current prescribing practice. The combination of AI drug design and AI-driven patient stratification promises more effective treatments with fewer side effects.

The risks are real. AI-designed molecules are novel, and their long-term safety profiles are untested. The compression of discovery timelines means less time for unexpected side effects to surface during development. Patients and physicians should expect AI-discovered drugs to carry the same rigorous clinical trial requirements as traditional drugs, but with the potential benefit of reaching patients years earlier than would otherwise be possible.