This website presents our research on molecular representations, virtual screening and protein–ligand binding affinity. The articles explain UniMolRep, DL-FSG and AGIMA-Score alongside figures and findings from our papers.
You can explore how molecular structures become inputs for learning models and how those models make predictions. Interactive analyses let you compare representations, try a screening model and examine the spatial response of an affinity model.
UniMolRep · 2026
A fingerprint, a graph and a spatial grid expose different information to a model. UniMolRep makes these choices explicit and brings their preparation into one toolkit.
Its benchmarks examine where complementary representations improve molecular-property and affinity prediction.
Explore representations →DL-FSG · 2026
Our screening study connects a ligand’s chemical sequence with its three-dimensional grid through cross-attention. The DUD-E evaluation tests the contribution of each representation and their interaction.
Explore screening →AGIMA-Score · 2025
Which atomic connections matter when estimating binding strength? AGIMA-Score builds graphs around contacts between protein and ligand atoms, then evaluates this choice on independent scoring tests.
Explore affinity prediction →The project investigates how molecular structure can be represented and learned to estimate protein–ligand binding affinity. It aims to develop grid representations and deep-learning methods for affinity prediction and rescoring in structure-based drug design.