From Data to Drugs: The Role of Artificial Intelligence in Drug Discovery

Abstraction: AI and foundation models transforming pharmaceutical drug discovery pipelines

Key points:

• Traditional drug discovery takes 13–15 years with less than 10% of Phase I candidates receiving FDA approval and average R&D cost exceeding $2.5 billion
• AlphaFold (DeepMind) sparked a wave of biological foundation models including RoseTTAFold and Meta's ESM family; Isomorphic Labs secured partnerships with Eli Lilly and Novartis worth ~$3B
• Amgen open-sourced AMPLIFY, a protein language model demonstrating that data quality can surpass model size while achieving state-of-the-art protein sequence prediction
• Reinforcement learning used to fine-tune compound properties like synthesizability and drug-likeness within enormous chemical search spaces
• Generative AI often proposes compounds that are difficult to synthesize or lack drug-like properties; benchmarking via MoleculeNet has known flaws; Polaris platform (backed by Merck, Pfizer, AstraZeneca, J&J, Novartis, Bayer) aims to improve benchmarking
• FDA Modernization Act 2.0 (Dec 2022) opened door for non-animal preclinical testing including organoids and organs-on-chips

Connections: Alphafold · Deepmind · Amgen · Isomorphic Labs · Drug Discovery · Foundation Models · Protein Structure Prediction · Reinforcement Learning

Source: https://www.linkedin.com/pulse/from-data-drugs-role-artificial-intelligence-drug-milad-alucozai-exmme/