Advancing Ligand-based Virtual Screening and Molecular Generation with Pretrained Molecular Embedding Distance
We propose pretrained embedding distance (PED), a molecular similarity measure computed directly from pretrained models without task-specific training, effective for both ranking molecules in virtual screening and guiding molecular generation via reward design.
Apr 27, 2026
Advancing Drug Discovery with Enhanced Chemical Understanding via Asymmetric Contrastive Multimodal Learning
We propose an asymmetric contrastive multimodal learning framework, an effective and training-efficient framework tailored for molecules, promoting cross-modality understanding between the molecular graph and other chemical modalities.
Jul 23, 2025
Human-Knowledge Integrated Methods for Learning With Small-Data Challenges
My doctoral dissertation on integrating human and expert knowledge into representation learning to address small-data challenges in biomedical and scientific applications.
Feb 1, 2025
Motif-based graph representation learning with application to chemical molecules
We designed a novel convolution module for graph representational learning on molecules with an efficient pretraining strategy, enabling the capture of local structural and semantic information from graph motifs.
Jan 11, 2023