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

Yinuo Zhang, Divya Srijay, Pranam Chatterjee

Track: Machine learning: computational method and/or computational results

Nature Biotechnology: Yes

Keywords: Conditional diffusion, Phage display, Heavy Metals

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

Abstract: We present Metalorian, a conditional diffusion model tailored to generate de novo heavy metal-binding peptides. Our approach leverages the embedding space of MetaLATTE, a multi-label classifier fine-tuned on known metal-binding sequences, to guide the generation of peptides with specific metal-binding capabilities. The model utilizes a co-evolving diffusion framework that simultaneously handles continuous protein embeddings and discrete metal-binding properties, allowing for focused generation of shorter, economically-viable peptides. We demonstrate the effectiveness of our approach by generating peptide binders for copper, cadmium, and cobalt binding. Our results show that the generated peptides maintain key properties such as charge and hydrophobicity while significantly reducing sequence length and molecular weight compared to known metal-binding proteins. Co-folding and binding energy analysis using molecular dynamics further validate the potential binding capacities of these novel sequences. Finally, we experimentally demonstrate that Metalorian-generated peptides effectively bind to cobalt resin via phage display. Overall, our work solidifies a foundational platform for designing heavy metal-binding peptides for targeted bioremediation campaigns, and further motivates utilization of well-trained, continuous latent spaces for diffusion-based de novo peptide design.

Anonymization: This submission has been anonymized for double-blind review via the removal of identifying information such as names, affiliations, and identifying URLs.

Presenter: ~Yinuo_Zhang3

Format: Yes, the presenting author will definitely attend in person because they attending ICLR for other complementary reasons.

Funding: No, the presenting author of this submission does *not* fall under ICLR’s funding aims, or has sufficient alternate funding.

Submission Number: 11