ProjFusNet: deep neural network for peptide precursor prediction using projection-fused protein language model and...

Area of Science:

• Biochemistry
• Bioinformatics
• Computational Biology

Background:

• Peptide precursors are vital for neuroregulation, immune defense, and drug development.
• Accurate identification of peptide precursors is essential for understanding biological mechanisms and creating new therapies.
• Current prediction methods struggle with the complexity of peptide sequences, limiting performance.

Purpose of the Study:

• To develop a deep learning framework for improved peptide precursor identification.
• To leverage complementary sequence and structural information for enhanced prediction accuracy.
• To overcome limitations of existing methods that rely on single feature types.

Main Methods:

• Introduced ProjFusNet, a deep learning framework integrating ESM-2 sequence representations and structural features.
• Employed a projected multimodal fusion strategy to combine diverse data modalities.
• Utilized a bidirectional LSTM to model intricate sequence-structure interactions.

Main Results:

• ProjFusNet demonstrated superior performance in peptide precursor prediction.
• The model showed significant improvements across key metrics like ACC, SN, AUC, SP, and MCC.
• Compared to single-feature models, ProjFusNet achieved higher accuracy by integrating multimodal data.

Conclusions:

• ProjFusNet offers a powerful new approach for peptide precursor identification.
• Integrating evolutionary sequence and structural data enhances prediction performance.
• This framework holds promise for advancing neuroregulation research and therapeutic development.