DeepLigand

DeepLigand enhances prediction of major histocompatibility complex (MHC) class I ligands to improve selection of peptides for peptide-based therapeutic design.

Key Features:

• Semi-Supervised Model: Employs a semi-supervised approach that integrates a peptide language model with traditional peptide binding affinity prediction to assess MHC class I peptide presentation.
• Peptide Language Model: Incorporates a peptide language model that captures sequence features relevant to secondary factors influencing MHC ligand selection, including aspects of the peptide display pathway.
• Pre-trained Peptide Embedding: Uses pre-trained peptide embeddings trained on natural ligands to discriminate ligands from non-ligands even when binding affinity data are absent.
• High Predictive Accuracy: Demonstrates high accuracy in distinguishing ligands from non-ligands across a range of binding affinities, including peptides with moderate affinities.

Scientific Applications:

• Peptide-based therapeutic design: Facilitates identification of candidate peptides for peptide-based therapeutics by combining sequence-derived features with binding affinity information.
• Personalized medicine: Supports personalized medicine efforts by improving selection of patient-specific MHC class I ligands.
• Vaccine development: Aids vaccine development by enhancing prediction of MHC class I ligands relevant to vaccine antigen selection.

Methodology:

Combines semi-supervised machine learning that integrates a peptide language model with traditional peptide binding affinity prediction and pre-trained embeddings derived from natural ligands.