| Property | Value |
|---|---|
| Author | ChatterjeeLab |
| Model Type | Masked Language Model |
| Size | 650M parameters |
| Access | Gated (requires approval) |
| Platform | HuggingFace |
What is PepMLM-650M?
PepMLM-650M is an innovative AI model designed for the de novo generation of linear peptide binders. Built on the ESM-2 protein language model architecture, it employs a unique masking strategy to generate peptide sequences that can bind to specific target proteins without requiring structural information.
Implementation Details
The model implements a sophisticated masking approach that positions cognate peptide sequences at the terminus of target protein sequences. It utilizes the ESM-2 architecture to reconstruct the binder region, achieving competitive perplexity scores on validated peptide-protein sequence pairs.
- Built on ESM-2 protein language model architecture
- Novel terminal masking strategy for peptide generation
- Validated using AlphaFold-Multimer
- Experimental verification through E3 ubiquitin ligase domain fusion
Core Capabilities
- De novo generation of peptide binders
- Target sequence-conditioned design
- Structure-independent binding prediction
- Support for programmable proteome editing
- Endogenous protein degradation validation
Frequently Asked Questions
Q: What makes this model unique?
PepMLM-650M stands out for its ability to generate peptide binders without requiring target protein structure information, using only sequence data. This makes it more accessible and widely applicable compared to structure-based approaches.
Q: What are the recommended use cases?
The model is particularly suited for protein engineering applications, drug design, and programmable proteome editing. It's especially valuable when structural data is unavailable or when rapid screening of potential peptide binders is needed.