Morphology-Aware Peptide Discovery via Masked Conditional Generative Modeling.

Area of Science:

• Biomaterials Science
• Computational Chemistry
• Peptide Design

Background:

• Peptide self-assembly is key for creating biocompatible materials.
• Predicting and controlling peptide aggregate morphology is challenging due to vast sequence space.

Purpose of the Study:

• To develop an end-to-end pipeline, PepMorph, for designing peptides with predictable self-assembly morphologies.
• To steer peptide self-assembly towards fibrillar or spherical structures using conditional descriptors.

Main Methods:

• Compiled a dataset of peptide aggregation propensity and descriptors.
• Trained a Transformer-based Conditional Variational Autoencoder (CVAE) for peptide generation.
• Validated generated sequences using coarse-grained molecular dynamics (CG-MD) simulations.

Main Results:

• PepMorph successfully generates peptide sequences prone to aggregation.
• The pipeline steers self-assembly towards desired morphologies (fibrillar or spherical).
• Achieved an 83% success rate in CG-MD validation for targeted morphology classes.

Conclusions:

• PepMorph provides a robust framework for application-driven peptide discovery.
• The method enables controlled design of peptide self-assembly for biomaterial applications.
• Conditional generation based on peptide descriptors is effective for morphology control.