Metalorian: De Novo Generation of Heavy Metal-Binding Peptides with Classifier-Guided Diffusion Sampling

Yinuo Zhang, Divya Srijay, Pranam Chatterjee

Track: Full Paper Track

Keywords: Metal binder generation, guided diffusion sampling

TL;DR: We develop a novel conditional diffusion model on pLM latents to generate and experimentally validate heavy metal-binding peptides.

Abstract: Metalorian is a conditional diffusion model for de novo generation of heavy metal-binding peptides. By leveraging MetaLATTE’s embedding space, a multi-label classifier fine-tuned on known metal-binding sequences, our model guides peptide generation with specific metal-binding capabilities. Using a co-evolving diffusion framework, Metalorian jointly optimizes continuous protein embeddings and discrete metal-binding properties, enabling the design of shorter, cost-effective peptides. We generate and validate binders for copper, cadmium, and cobalt, demonstrating that the peptides retain key properties such as charge and hydrophobicity while reducing sequence length and molecular weight. Molecular dynamics confirm potential binding capacity, and phage display experiments validate cobalt binding. Our work provides a scalable platform for designing metal-binding peptides for bioremediation and highlights the utility of structured latent spaces in diffusion-based peptide design.

Attendance: Yinuo Zhang, Divya Srijay, Pranam Chatterjee

Submission Number: 26