[Paper Review] PepMLM: Target Sequence-Conditioned Generation of Therapeutic Peptide Binders via Span Masked Language Modeling

Tianlai Chen, Dumas, Madeleine+18 more|PubMed|Oct 5, 2023

Protein Degradation and Inhibitors29 references17 citations

TL;DR

PepMLM is a target-sequence conditioned generator that designs de novo peptide binders using a span masking strategy and fine-tunes a protein language model, with in silico and experimental validation of binder efficacy.

ABSTRACT

Target proteins that lack accessible binding pockets and conformational stability have posed increasing challenges for drug development. Induced proximity strategies, such as PROTACs and molecular glues, have thus gained attention as pharmacological alternatives, but still require small molecule docking at binding pockets for targeted protein degradation. The computational design of protein-based binders presents unique opportunities to access "undruggable" targets, but have often relied on stable 3D structures or structure-influenced latent spaces for effective binder generation. In this work, we introduce <b>PepMLM</b>, a target sequence-conditioned generator of <i>de novo</i> linear peptide binders. By employing a novel span masking strategy that uniquely positions cognate peptide sequences at the C-terminus of target protein sequences, PepMLM fine-tunes the state-of-the-art ESM-2 pLM to fully reconstruct the binder region, achieving low perplexities matching or improving upon validated peptide-protein sequence pairs. After successful <i>in silico</i> benchmarking with AlphaFold-Multimer, outperforming RFDiffusion on structured targets, we experimentally verify PepMLM's efficacy via fusion of model-derived peptides to E3 ubiquitin ligase domains, demonstrating endogenous degradation of emergent viral phosphoproteins and Huntington's disease-driving proteins. In total, PepMLM enables the generative design of candidate binders to any target protein, without the requirement of target structure, empowering downstream therapeutic applications.

Motivation & Objective

• Address how to generate de novo peptide binders for targets lacking accessible binding pockets or stable structures.
• Develop a target sequence-conditioned generator that places cognate binder sequences at the C-terminus for effective reconstruction.
• Evaluate the approach with AlphaFold-Multimer benchmarking and experimental degradation assays.
• Enable binder design for any target protein without relying on target structure.

Proposed method

• Introduce span masking that positions the cognate binder sequence at the target’s C-terminus.
• Fine-tune the state-of-the-art ESM-2 protein language model to reconstruct the binder region.
• Perform in silico benchmarking against AlphaFold-Multimer and compare with RFDiffusion on structured targets.
• Experimentally validate generated binders by fusing them to E3 ligase domains to induce degradation of target proteins.
• Assess the ability to generate candidate binders without requiring target structure knowledge.

Experimental results

Research questions

• RQ1 Can a span masking strategy conditioned on a target sequence enable reliable generation of binder regions?
• RQ2 Does fine-tuning a protein language model on this conditioning improve binder reconstruction and plausibility?
• RQ3 How does PepMLM perform in silico relative to existing diffusion-based approaches on structured targets?
• RQ4 Are PepMLM-derived binders effective in driving degradation of target proteins in experimental assays?

Key findings

• PepMLM achieves low perplexities for binder reconstruction that match or improve upon validated binder pairs.
• In silico benchmarking with AlphaFold-Multimer supports PepMLM’s effectiveness and outperforms RFDiffusion on structured targets.
• Experimental validation demonstrates endogenous degradation of emergent viral phosphoproteins and Huntington’s disease–driving proteins using model-derived peptides fused to E3 ligase domains.
• The approach enables generation of candidate binders for any target protein without requiring target structure.