TideFrag-GNN: A Fragment-Graph Foundation Encoder for Medium-Sized Peptide Property Prediction

Overview

Oral presentation at the CBI Society Annual Meeting 2026 on TideFrag-GNN, a fragment-graph encoder that decomposes peptide SMILES into monomer-scale fragments and pretrains a GraphGPS model on them, so that non-natural monomers outside a predefined library can still be represented. It extends the earlier HELM-BERT work. This presentation received the Excellent Presentation Award.

Session: O09 “AI Drug Discovery” (O09-06, 10:00–11:30)

Affiliation: Kyoto University Graduate School of Medicine, Department of Biomedical Data Intelligence