Accelerate Peptide Binder Design: PepMLM Now Available on Vecura

What is PepMLM?

PepMLM (Peptide Masked Language Model) is a generative AI model designed to create linear peptide binders conditioned exclusively on the amino acid sequence of a target protein. By leveraging the deep evolutionary knowledge embedded within the 650-million-parameter ESM-2 protein language model, it reconstructs potential peptide binders through iterative masked token sampling. It is especially useful for researchers looking to rapidly identify candidate ligands for protein-protein interactions or E3 ligase recruitment in targeted protein degradation without requiring computationally intensive 3D structural data.

What can users do with PepMLM on Vecura?

With PepMLM on Vecura, users can:

• Generate novel linear peptide binders by inputting only a target protein's amino acid sequence.
• Customize peptide length (1–50 residues) to tailor the design to specific binding pockets or functional requirements.
• Receive a prioritized list of candidates ranked by their pseudo-perplexity score, which predicts the model's confidence in the peptide-target interaction.
• Streamline early-stage drug discovery by eliminating the need to set up complex local compute infrastructure or perform costly 3D docking simulations.

What the output means

The output provides a ranked list of candidate peptide sequences, each accompanied by a pseudo_perplexity score. A lower score signifies that the language model assigns a higher probability to the peptide being a cognate binder for the given target sequence.

This output should be used to support scientific decision making by helping to prioritize candidates for subsequent wet-lab validation. It does not replace experimental validation, as pseudo-perplexity is a proxy for affinity rather than a calibrated binding constant.

Why this matters

The design of peptide binders has traditionally relied on either heavy experimental screening or computationally expensive structure-based methods that demand precise 3D coordinates, which are not always available for novel targets. PepMLM shifts this paradigm by enabling purely sequence-conditioned generation, drastically lowering the barrier to entry for initial binder discovery.

By accelerating the identification of high-probability binder candidates, this tool allows researchers to focus experimental resources on the most promising sequences. This is particularly valuable for accelerating the development of novel therapeutic modalities, such as PROTAC-style targeted protein degraders, where efficient recruitment of E3 ligases is a critical hurdle.

• Developed by: Lissovsky et al.
• Source: Official GitHub repository
• Reference: Lissovsky et al., "PepMLM: Target Sequence-Conditioned Generation of Peptide Binders via Masked Language Modeling", arXiv 2023; Nature Biotechnology (2025)