AstraZeneca

AstraZeneca needed to identify viable drug targets from massive genomic datasets efficiently. Traditional case-control studies risked misclassifying individuals as controls, reducing statistical power for genetic discovery. The company also sought to overcome confirmation bias in conventional target identification methods.

AstraZeneca built a Center for Genomics Research and developed MILTON (MachIne Learning with phenoType associatiONs), an ML tool trained on ~500,000 UK Biobank participants to predict diseases before diagnosis. They also collaborated with BenevolentAI to use knowledge graphs integrating genomic, disease, drug, clinical, and safety data to identify novel drug targets.

MILTON achieved high predictability (AUC > 0.7) for 1,091 diseases and exceptional performance (AUC > 0.9) for 121 diseases. AI-driven reclassification expanded gene discovery scope across hundreds of diseases. AstraZeneca added the first two AI-generated drug targets to their portfolio through the BenevolentAI collaboration in 2021.

• Large genomic datasets combined with ML can predict diseases pre-diagnosis with clinical-grade accuracy across hundreds of conditions.
• Knowledge graphs synthesizing genomic, clinical, and safety data can surface drug targets that confirmation-bias-prone traditional methods miss.
• Partnering with specialist AI vendors (BenevolentAI) can accelerate the initial pipeline of AI-generated target candidates.