Anticancer peptides prediction with deep representation learning features.

Area of Science:

• Computational biology
• Bioinformatics
• Peptide therapeutics

Background:

• Anticancer peptides are promising cancer therapeutics.
• Experimental verification of anticancer peptides is resource-intensive.
• Need for efficient computational methods for anticancer peptide identification.

Purpose of the Study:

• To develop a computational method (iACP-DRLF) for identifying anticancer peptides.
• To leverage deep representation learning features for enhanced prediction accuracy.
• To compare different sequence embedding techniques for anticancer peptide classification.

Main Methods:

• Proposed iACP-DRLF method using light gradient boosting machine.
• Employed deep representation learning with soft symmetric alignment and UniRep embeddings.
• Utilized deep neural network models based on long short-term memory networks.

Main Results:

• Deep representation learning features significantly improved model discrimination of anticancer peptides.
• UniRep embedding demonstrated an advantage over other features in identification.
• UMAP and SHAP analyses validated the effectiveness of UniRep features.

Conclusions:

• The iACP-DRLF method effectively identifies anticancer peptides using deep learning features.
• UniRep embedding offers superior performance for anticancer peptide prediction.
• Computational approaches accelerate the discovery of novel anticancer peptide therapeutics.