Institute of Information Science, Academia Sinica

Peptide nanofibrils are versatile biomaterials built from elementary peptide molecules, periodic assembly patterns, and supramolecular geometries. One key factor in stabilising peptide nanofibril structures is peptide-peptide interactions. These peptide-peptide interactions are usually complex and involve massive noncovalent bonding. Uncovering reduced variables for generating peptide nanofibrils is therefore desirable. In our recent work, we propose geometrical parameters for assembling peptides into fibrils and use molecular dynamics simulations to collect energetically favourable configurations. We identify various periodic peptide packings to support fibril structures (Yang, 2025). We further locate supramolecular geometries that properly arrange peptide-peptide interactions (Yang, 2026). Our simulated data closely align with experimental observations, providing detailed noncovalent bonding information and resolving energetically favourable fibril structures. Next, we explore how peptide molecules can affect peptide-peptide interactions and shape their assembly behaviours. Alongside deep generative models that relate biomolecules to their structural ensembles (Abramson, 2024; Lewis, 2025; Rehman, 2026), we will discuss how data integration, model architectures, and objective functions can influence model performance. Robust, reliable structural inference could help design peptide molecules and the nanomaterials they assemble into.