Seminar| Institute of Mathematical Sciences
Time: Wednesday, June 24th, 2026,13:45-14:30
Location: IMS R307
Speaker: Shuhui Wang, Université Paris Dauphine-PSL
Abstract:Artificial intelligence is fundamentally reshaping drug discovery by tackling three critical challenges: predicting drug safety, uncovering synergistic combinations, and simulating cellular responses.
ADMET Prediction addresses the leading cause of clinical failure—poor pharmacokinetics and unforeseen toxicity. Traditional experimental screening is slow and costly, but AI models—particularly graph neural networks and multitask learning—now predict Absorption, Distribution, Metabolism, Excretion, and Toxicity directly from molecular structure. These models learn complex structure–property relationships, enabling rapid virtual screening and lead optimization while reducing late-stage attrition.
Synergistic Drug Discovery tackles diseases like cancer where single drugs are insufficient. Exhaustive experimental testing of all drug pairs is infeasible. Deep learning frameworks integrate multimodal data—genomic profiles, drug perturbation signatures, and knowledge graphs—to predict combination efficacy with high accuracy.
Virtual Cell Perturbation represents the frontier—modeling how genetic or chemical interventions alter cellular states. Foundation models trained on massive transcriptomic datasets can now predict responses to unseen compounds across different cell types, from simple lines to complex organoids. These virtual cells allow researchers to test hypotheses in silico, dramatically reducing reliance on costly experiments.
Together, these domains form an integrated AI pipeline: ADMET ensures candidates are safe, synergy identifies optimal pairings, and virtual cells predict real-world biological outcomes. This paradigm shifts drug discovery from empirical trial-and-error to a predictive science—accelerating timelines, cutting costs, and increasing clinical success rates. While challenges in interpretability and generalization remain, AI is steadily turning drug development into a digitally guided, data-driven discipline.