Developing a new medicine typically takes many years and enormous sums, and most candidates fail along the way. Artificial intelligence promises to make each step faster and more successful. Some of that promise has already been realized — most famously in protein structure prediction — while other claims are still being tested in clinical trials.

The breakthrough: protein structures

A protein’s 3D shape determines its function and how drugs can bind to it. Determining structures experimentally can take months or years. In 2020, Google DeepMind’s AlphaFold 2 predicted structures with accuracy comparable to experimental methods in the CASP14 competition, and the method was published in Nature in 2021.

DeepMind and the European Bioinformatics Institute then released the AlphaFold Protein Structure Database, with predicted structures for hundreds of millions of proteins — nearly all catalogued proteins known to science. In 2024, Demis Hassabis and John Jumper shared the Nobel Prize in Chemistry for protein structure prediction, together with David Baker for computational protein design.

Later systems, including AlphaFold 3 (2024), extended prediction to how proteins interact with DNA, RNA and small molecules — directly relevant to drug design.

Where AI is used across the pipeline

  1. Target identification: analyzing genetic, clinical and molecular data to find which proteins or pathways to target.
  2. Hit discovery: virtually screening huge libraries of molecules to predict which might bind a target.
  3. Generative chemistry: designing new molecules with desired properties rather than only searching existing ones.
  4. Optimization: predicting toxicity, solubility and how the body will process a compound, reducing failed experiments.
  5. Protein and antibody design: designing biologic drugs and enzymes computationally.
  6. Clinical trials: identifying suitable patients, choosing trial sites and analyzing results.

What has reached the clinic

Several drug candidates discovered or designed with significant AI involvement have entered human clinical trials. Some companies report reaching clinical candidates faster and with fewer synthesized molecules than traditional approaches. A few AI-discovered candidates have reported mid-stage trial results. However, no drug has yet been approved whose discovery is attributed mainly to AI, and the key question — whether AI improves the success rate in trials, not just speed in early stages — is still being answered.

Why it is hard

  • Biology is complex: a molecule that binds well in silico may fail in cells, animals or people.
  • Data scarcity and quality: many datasets are small, noisy or proprietary.
  • Validation takes time: clinical trials still take years regardless of how fast discovery is.
  • Hype cycles: early announcements often outpace evidence.

Partnerships and business models

Large pharmaceutical companies partner with AI-first biotech firms and technology companies, combining computational methods with laboratory and clinical expertise. Google DeepMind spun out Isomorphic Labs in 2021 to apply its methods to drug discovery.

How a typical AI-assisted program works

A simplified example shows how the pieces fit together. A team targeting a protein involved in a disease might:

  1. use genetic and clinical data to confirm the protein matters in the disease;
  2. use a predicted or experimental structure to find pockets where a small molecule could bind;
  3. virtually screen millions or billions of compounds and use generative models to propose new ones;
  4. synthesize and test a few hundred of the most promising molecules, rather than many thousands;
  5. feed lab results back into the models in repeated “design–make–test–analyze” cycles;
  6. predict and test safety and pharmacokinetic properties to choose a clinical candidate.

The key advantage is the tight loop between prediction and experiment. AI does not remove the laboratory; it makes each round of experiments more informative.

What to watch next

  • Phase 2 and 3 results for AI-discovered drug candidates.
  • The first regulatory approval of a drug discovered largely with AI.
  • Open versus proprietary access to new structure-prediction models.
  • Regulators’ guidance on AI used in drug development and trials.

Key takeaways

  • AlphaFold transformed protein structure prediction and won a share of the 2024 Nobel Prize in Chemistry.
  • AI supports target discovery, screening, molecule design, optimization and trials.
  • Several AI-discovered candidates are in clinical trials; none is yet approved as a mainly AI-discovered drug.
  • Speed gains are clearer than improvements in clinical success rates so far.
  • Biology’s complexity, data limits and long trials temper expectations.

Frequently asked questions

Has AI discovered an approved drug?

Not yet in the sense of a drug discovered mainly by AI, though AI tools have contributed to many programs. Several AI-originated candidates are in clinical trials.

Is the AlphaFold database free?

Yes, the AlphaFold Protein Structure Database is freely accessible for research and other uses under its terms.

Can AI replace lab experiments?

No. AI narrows down options and predicts properties, but experiments and clinical trials remain essential to confirm safety and efficacy.