Drug discovery is one of the fastest-moving areas of AI commercialization in healthcare. The industry-recognized problem: traditional development from target to approval takes 10–15 years at a cost of $1–2 billion, with roughly 90% of clinical candidates ultimately failing to achieve approval.
## AI Applications Across Development Stages
**Target identification**: AI analyzes large-scale genomic, proteomic, and disease datasets to identify druggable targets. Biological validation of targets still requires wet lab work — AI predictions need experimental confirmation.
**De novo molecular design**: generative AI models (graph neural networks, diffusion models) generate molecules with specified properties (binding affinity, selectivity, drug-likeness) from scratch. Insilico Medicine’s INS018_055 — an AI-generated candidate for idiopathic pulmonary fibrosis — entered Phase II clinical trials in 2023, marking a significant milestone for AI-designed molecules reaching clinical development.
**ADMET property prediction**: AI models predict absorption, distribution, metabolism, excretion, and toxicity properties computationally, substantially reducing early-stage animal testing requirements.
**Protein structure prediction**: DeepMind’s AlphaFold 2 (2021) and AlphaFold 3 (2024) accurately predict three-dimensional protein structures for essentially all human proteins, fundamentally transforming structural biology and directly accelerating structure-based drug design. The AlphaFold database is publicly available at [alphafold.ebi.ac.uk](https://alphafold.ebi.ac.uk/).
**Clinical trial optimization**: AI identifies eligible patients through EMR data, selects biomarker-defined patient populations most likely to benefit, and optimizes adaptive trial designs.
## Representative Companies
Insilico Medicine (generative AI, two molecules in clinical trials), Recursion Pharmaceuticals (phenotypic cell imaging at scale), Schrödinger (physics-based computational chemistry), XtalPi (crystal form prediction, listed), Exscientia (AI-first drug design), Generate:Biomedicines (protein generative AI).
## Realistic Expectations
AI drug discovery is in early validation. Entering clinical trials does not equal eventual approval — most candidates still fail clinically. AI’s contribution is clearer at: reducing early screening time and cost, identifying higher-quality starting molecules, and providing specific efficiency gains in tasks like ADMET prediction and crystal structure analysis. See [AI Medical Diagnosis](https://sunqi.org/ai-medical-diagnosis-tools-en/).
