Loading article...
Article
Loading article...
Discover which TRIZAN solutions align with your goals using our interactive Solution Finder—results in 3 minutes.
Discover Your TRIZAN PlanFor fifty years, the protein structure prediction problem stood as one of the central unsolved challenges in structural biology: given a protein's amino acid sequence, predict the three-dimensional structure into which it folds. Knowing a protein's structure is essential for understanding its biological function and for designing molecules that interact with it — the foundation of rational drug design. But experimental methods for determining protein structure — X-ray crystallography, cryo-electron microscopy, NMR spectroscopy — are expensive, slow, and fail for a substantial proportion of proteins. The scientific community had accumulated sequence data for hundreds of millions of proteins but structural data for only a tiny fraction. In November 2020, DeepMind's AlphaFold 2 achieved prediction accuracy comparable to experimental methods on the Critical Assessment of Protein Structure Prediction (CASP) benchmark, essentially solving the problem. In July 2021, DeepMind released AlphaFold 2 as open source and published the predicted structures of the entire human proteome — approximately 20,000 proteins. By 2022, it had released predicted structures for over 200 million proteins across nearly all known organisms. The structural biology data that had taken the field decades to accumulate at enormous cost became available for the entire known protein universe in 18 months. The Drug Discovery Acceleration: From Decades to Years The traditional pharmaceutical drug discovery timeline — from initial target identification to approved drug — averages 12 to 15 years and costs $1 to $2 billion, with a failure rate exceeding 90% across clinical trial phases. This timeline and cost structure are primarily consequences of the information problem: understanding which molecular targets are involved in a disease, finding compounds that interact with those targets effectively, predicting and managing off-target effects and toxicity, and navigating the regulatory demonstration of safety and efficacy all involve enormous amounts of trial and error against a background of incomplete biological information. AI is compressing this timeline through improvements at each stage of the discovery and development process. AlphaFold and its successors (RoseTTAFold, ESMFold, and the more recent generation of structure prediction models) have transformed target identification and structure-based drug design by making high-resolution structural data available for virtually any biological target. Generative AI models trained on molecular structure data can now design novel drug candidates with specified binding affinity and selectivity profiles — reversing the traditional discovery approach of screening libraries of existing compounds and instead designing new molecules optimized for a specific target from scratch. AI models trained on clinical trial data and electronic health records can predict trial failure modes, optimize patient selection criteria, and identify biomarkers that distinguish responders from non-responders with significantly higher accuracy than conventional trial design. The commercial results of AI-accelerated drug discovery are beginning to appear in clinical pipelines. Insilico Medicine received FDA approval in 2023 for an IND application on a drug designed entirely by AI — a fibrosis drug that went from target identification to clinical trial candidate in 18 months, compared to the 4-6 year average for the equivalent conventional process. Recursion Pharmaceuticals, BioNTech (applying its mRNA