MultiPep-DLCL: recognition of multifunctional therapeutic peptides through deep learning with label-sequence...

Area of Science:

• Computational Biology
• Bioinformatics
• Drug Discovery

Background:

• Identifying multifunctional therapeutic peptides (MFTPs) is crucial but challenging due to complex labeling requirements.
• Current methods often overlook the detailed semantics of amino acid labels and the sequence-label interplay.
• Accurate MFTP classification is vital for advancing peptide-based therapeutics.

Purpose of the Study:

• To develop an advanced deep learning model for accurate MFTP classification.
• To address the limitations of existing methods in capturing nuanced label information and sequence-label interactions.
• To improve the recognition of multifunctional therapeutic peptides.

Main Methods:

• Proposed MultiPep-DLCL, a deep learning architecture for MFTP classification.
• Utilized a Label-Sequence Fusion Transformer to learn high-quality label embeddings from peptide sequences.
• Employed label-sequence contrastive learning to strengthen feature correspondence.
• Integrated a multilabel focal dice loss function to handle dataset imbalance.

Main Results:

• MultiPep-DLCL demonstrated superior performance in MFTP recognition compared to existing methods.
• The model effectively learned local and global dependencies within peptide sequences.
• High-quality label embeddings were successfully mined by leveraging peptide sequence information.
• The proposed loss function effectively addressed challenges posed by imbalanced datasets.

Conclusions:

• MultiPep-DLCL offers a significant advancement in the field of multifunctional therapeutic peptide recognition.
• The method's ability to integrate sequence features and label embeddings provides a robust framework for complex peptide classification.
• This approach holds promise for accelerating the discovery and development of novel peptide therapeutics.