Deep2Pep: A deep learning method in multi-label classification of bioactive peptide.

Area of Science:

• Computational Biology
• Peptide Science
• Machine Learning in Drug Discovery

Background:

• Functional peptides offer therapeutic potential due to easy absorption and low side effects.
• Screening large peptide libraries is hindered by resource and funding limitations.
• Machine learning and deep learning provide computational solutions for identifying peptide functions.

Purpose of the Study:

• To develop a deep learning model, Deep2Pep, for predicting multiple active peptide functions.
• To explore the potential of multi-functional active peptides in pharmaceutical research.

Main Methods:

• Deep2Pep utilizes sequence encoding, embedding, and language tokenization.
• The model integrates BiLSTM, attention-residual algorithm, and BERT Encoder for function prediction.
• Peptide sequences are converted into digital vectors for analysis.

Main Results:

• Deep2Pep achieved a Hamming Loss of 0.095, subset Accuracy of 0.737, and Macro F1-Score of 0.734.
• The model demonstrated superior performance compared to existing methods.
• BiLSTM was identified as the primary component, with BERT Encoder playing an auxiliary role.

Conclusions:

• Deep learning, specifically Deep2Pep, can accurately predict four key peptide functions: antimicrobial, antihypertensive, antioxidant, and antihyperglycemic.
• This approach offers a valuable reference for predicting multi-functional peptides efficiently.
• Deep2Pep facilitates the exploration of novel peptide therapeutics.