EnAMP: A novel deep learning ensemble antibacterial peptide recognition algorithm based on multi-features.

Area of Science:

• Computational biology
• Bioinformatics
• Drug discovery

Background:

• Antimicrobial peptides (AMPs) are crucial alternatives to antibiotics due to rising resistance.
• Experimental identification of AMPs is costly, time-consuming, and challenging.
• Machine learning (ML) approaches show promise for predicting AMPs.

Purpose of the Study:

• To develop an accurate and efficient ensemble ML model for AMP prediction.
• To integrate diverse feature types for improved predictive performance.
• To provide a publicly available tool for AMP identification.

Main Methods:

• Developed EnAMP, an ensemble classifier combining deep neural networks and traditional ML models.
• Utilized Word2vec and GloVe word embeddings for peptide sequence representation.
• Incorporated statistical features of peptide sequences.
• Averaged predictions from four distinct classifiers (two DNNs, Random Forest, SVM).

Main Results:

• EnAMP demonstrated superior performance compared to many state-of-the-art algorithms across six benchmark datasets.
• The model achieved comparable results to high-cost Bidirectional Encoder Representation from Transformers (BERT) models.
• EnAMP offers a favorable balance between computational cost and predictive accuracy.

Conclusions:

• EnAMP provides a robust and efficient computational method for identifying antimicrobial peptides.
• The ensemble approach effectively leverages different feature representations for enhanced prediction.
• EnAMP serves as a valuable tool for accelerating the discovery of novel AMPs.