Bristol Myers Squibb

Current approved medicines target only ~3% of human proteins, while approximately two-thirds of disease-relevant proteins lack potent binders and remain 'undruggable' with traditional approaches. Designing protein degrader molecules previously required labor- and time-intensive crystallography techniques, and the design-make-test-learn cycle for drug candidate nomination was slow and resource-intensive.

BMS integrated AI and machine learning across its entire protein degradation R&D pipeline: computer vision and deep learning neural networks analyze tissue images to extract clinically relevant features; pretrained foundation models built on single-cell and spatial transcriptomics data model gene regulatory networks and predict gene perturbation effects; geometric deep learning predicts protein complex structures for structure-based degrader design; diffusion-based generative AI designs new degrader molecules with drug-like properties; and closed-loop active learning workflows accelerate design-make-test-learn cycles for drug candidate nomination.

BMS describes accelerated timelines and enhanced capability to target previously 'undruggable' proteins through AI-enabled design. The company has advanced multiple investigational protein degrader modalities (molecular glues, LDDs, DACs) to clinical development, with early clinical validation already seen in multiple myeloma. Virtual simulation platforms are being used to predict and optimize trial success likelihood, and AI-based wearables enable real-time data collection to reduce patient burden.

• AI and machine learning are enabling pharma companies to expand the druggable proteome far beyond the ~3% currently targeted by approved medicines.
• End-to-end integration of AI — from target discovery through clinical trial design — compounds efficiency gains across the entire drug development pipeline.
• Generative AI and active learning workflows are qualitatively changing degrader molecule design speed, not just optimizing existing processes.