AI4Science Seminar #2 – Advancing Science with AI
Date
11 Jun 2026
AI4Science: A New Paradigm for Scientific Discovery
The second edition of the AI4Science Seminar series, hosted by SCAI, brought together researchers and experts at the intersection of artificial intelligence and scientific discovery.
A Multi-Task Deep-Learning Model for PPI Network Reconstruction and Interface Prediction" - by Sara Rescalli, PhD at CQSB.
Protein–protein interactions (PPIs) are central to cellular organization and regulation, shaping the molecular pathways that drive most biological processes. Yet experimental characterization of PPIs remains costly, time-consuming, and incomplete, leaving critical gaps in our knowledge of which proteins interact and where these interactions occur at the residue level.
Here we present X-PAIR, a multi-task deep learning framework that jointly predicts whether two proteins interact and, when they do, identifies the residues forming the interaction interface. X-PAIR operates from amino acid sequences alone, making it broadly applicable to the vast number of proteins for which structural data are unavailable.
We evaluate X-PAIR on established benchmarks for interaction and interface prediction, and further assess its generalization from an evolutionary perspective, testing whether predictions remain reliable across species phylogenetically distant from those seen during training.
Overall, X-PAIR demonstrates how AI can address challenging open problems in biology, extending interaction and interface annotation to organisms where experimental data and structural information remain limited.
We evaluate X-PAIR on established benchmarks for interaction and interface prediction, and further assess its generalization from an evolutionary perspective, testing whether predictions remain reliable across species phylogenetically distant from those seen during training.
Overall, X-PAIR demonstrates how AI can address challenging open problems in biology, extending interaction and interface annotation to organisms where experimental data and structural information remain limited.