Handcrafted versus non-handcrafted (self-supervised) features for the classification of antimicrobial peptides:...

Area of Science:

• Computational chemistry
• Bioinformatics
• Drug discovery

Background:

• Antimicrobial peptides (AMPs) are crucial for combating multi-drug resistant bacteria.
• Quantitative sequence-activity models (QSAMs) aid in discovering new AMPs by analyzing peptide sequences.
• Feature selection is critical for building effective QSAMs.

Purpose of the Study:

• To systematically compare handcrafted and non-handcrafted (unsupervised) features for AMP classification.
• To determine the optimal feature set for developing accurate QSAMs.
• To assess the performance of shallow learning models with different feature combinations.

Main Methods:

• Development and comparison of QSAMs using handcrafted and non-handcrafted features.
• Analysis of feature relevance and importance using statistical methods.
• Benchmarking shallow learning models against state-of-the-art deep learning models.

Main Results:

• Non-handcrafted features significantly outperform handcrafted features in AMP classification.
• Merging both feature types leads to improved model performance.
• Shallow models with non-handcrafted features, or combined features, outperform deep models.

Conclusions:

• Unsupervised approaches generate superior features for AMP QSAMs compared to handcrafted features.
• Complementarity exists between handcrafted and non-handcrafted features, enhancing model accuracy when combined.
• Shallow learning models offer a competitive alternative to deep learning for AMP discovery.