Prospectively validated augmented intelligence for disease-agnostic predictions of clinical success for novel therapeutics

Standalone artificial intelligence has not alleviated the long-term bottleneck of linearly extracting new knowledge from exponentially growing biological data, which has severely limited clinical success rates for drug discovery. We developed a ‘virtual patient’ augmented intelligence model that functionally reconstructed human physiology and human pathogenesis for high-fidelity simulations of drug-body interactions. We examined the clinical utility of ‘virtual patient’ in prospective predictions of clinical efficacy and safety of novel therapeutics regardless of prior clinical data availability, through a 24- month, public, prospective, large-scale, unbiased, and real-world validation study. ‘Virtual patient’ achieved 90.1% sensitivity and 82.0% precision with a 99% confidence across all major therapeutic areas, representing its capability of discovering 90.1% of all possible drug-indication pairs that could bring clinical benefits to patients, and its potential of increasing tenfold the baseline clinical success rate from 7.9% to 82.0%. ‘Virtual patient’ represents a methodological shift of drug discovery especially for age-related diseases by doing away with animal experiments whose data are hard to reproduce, virtualizing human trials whose outcomes are doomed to failure, initiating human trials whose participants are likely to benefit, and reducing R&D cycles and costs while increasing clinical efficacy and safety. One-Sentence Summary A prospectively validated ‘virtual patient’ achieved a 10.4-fold improvement in the clinical success rate for new drugs across all major diseases with 99% confidence.