Bioactivity prediction is a cornerstone of modern AI-driven drug discovery, enabling efficient identification of promising therapeutic candidates. However, existing approaches often suffer from three critical limitations: incomplete modeling of protein–ligand interactions, insufficient representation of binding site geometry, and poor generalization under domain shifts. In this talk, I will present a unified research framework addressing these challenges through three recent works. First, I introduce a Drug-Target Interaction Graph Neural Network (DTIGN) that integrates molecular docking with self-attention mechanisms to explicitly model intermolecular interactions, significantly improving bioactivity prediction accuracy. Second, I present LigoSpace, a novel approach that captures spatial emptiness within protein–ligand complexes and aggregates multiple binding pockets, offering a more holistic representation of binding environments beyond traditional local interaction modeling. Third, I discuss a practical test-time adaptation framework for out-of-domain bioactivity prediction that operates without access to source data, leveraging uncertainty-weighted consistency and contrastive learning to achieve robust generalization under realistic deployment constraints. Together, these works advance bioactivity prediction along three key dimensions: richer structural modeling, more expressive geometric representation, and stronger robustness to distribution shifts. This integrated perspective highlights a pathway toward reliable and deployable AI systems for real-world drug discovery.

Biography:

Dr. Adams Wai-Kin Kong received his Ph.D. degree from the University of Waterloo, Canada. Currently, he is an associate professor at Nanyang Technological University, Singapore, and the director of the Master of Science in Artificial Intelligence programme. His research works have been published in major AI conferences and journals. His papers were selected as spotlight papers by IEEE Transactions on Pattern Analysis and Machine Intelligence, ICML, and ICRL, and CVPR for oral presentation and received an Honorable Mention from Pattern Recognition. His PhD students received Best Student Paper Awards at The IEEE Fifth International Conference on Biometrics: Theory, Applications and Systems (BTAS), 2012, and the IEEE International Conference on Bioinformatics and Bioengineering, 2013, as well as other awards such as the Google Anita Borg Memorial Scholarship. He currently serves as an associate editor for IEEE Transactions on Biometrics, Behavior, and Identity Science, an editorial board member for Visual Computer, an area chair for major AI conferences such as NeurIPS, ICLR, CVPR, and ECCV, and an expert consultant for a cross-border legal case. Dr. Kong was nominated for the MSc in Business Analytics Teacher of the Year Award in 2022 and 2024. He received a faculty award from his school in 2022. In addition to teaching and academic research, he has been working on various industrial projects. His industrial partners include Rolls Royce, Siemens, Procter & Gamble, Hyundai, and Nanyang Biologics. His recent research interests include AI safety, deep networks and their applications in healthcare, drug discovery, computational simulation, and biometrics.