Work

Integrating Patient Centric Disease Models and AI-Assisted Mechanism Discovery for Translational Drug Repurposing

Year: 2026
Type: conference-paper
Abstract: Drug repurposing frequently fails at the transition from in vitro hit identification to clinical efficacy. We developed a phenotype-first drug repurposing platform built on authentic viral infections in human cell systems, high-content imaging, single-cell machine-vision phenotyping, and exposure-aware prioritization. Drug repurposing as a strategy for antiviral drug discovery has produced tractable results from viral high content assays including BK Polyomavirus, and SARS-CoV-2. BK Polyomavirus causes kidney injury in immunocompromised individuals and there is no approved therapy. Screening of 1,425 FDA-approved drugs yielded a 3.9% primary hit rate, with average Z′ = 0.70. 20 Hits were selected for acceptable cell-based therapeutic index (CC50/IC50) from efficacy and viability in the same wells, then triaged by PBPK-based reverse dosimetry, converting cellular IC50 to an equivalent human oral dose for comparison with approved-dose plasma exposure and NOAEL thresholds. Target deconvolution mechanistic imputation was performed with a chemically/biologically aware large language model integrating compound annotations and RNAseq data to generate plausible mechanistic hypotheses, and identifying promiscuous hits for deprioritization. Velpatasvir emerged as a post-entry, pan-polyomavirus lead: BKPyV IC50 429 nM, CC50 > 10 µM (SI > 23), 81 nM in a 7-day cytopathic-effect assay, and 212 nM against murine polyomavirus. The same platform nominated proxalutamide for Covid19/SARS-CoV-2 (IC50 97 nM); and in a randomized, double-blind, placebo-controlled trial in mild-to-moderate COVID-19, it reduced the overall risk of hospitalization or death by 50%, with a reported 100% reduction in high-risk patients, while also accelerating viral clearance and shortening time to clinical remission (NCT04870606, NCT05009732).
Open Access status: gold
OpenAlex ID: W7204240103
FWCI:
Cites:
Cited by:
Related to: