ACPred-BMF: bidirectional LSTM with multiple feature representations for explainable anticancer peptide prediction.

Area of Science:

• Biotechnology
• Computational Biology
• Drug Discovery

Background:

• Traditional cancer treatments like chemotherapy and radiotherapy have limitations, including low specificity and severe side effects.
• Anticancer peptides present a promising alternative due to their high efficacy, specificity, and low toxicity.
• Experimental identification of anticancer peptides is resource-intensive and not high-throughput.

Purpose of the Study:

• To develop a computational tool for accurate and efficient prediction of anticancer peptides.
• To overcome the limitations of experimental methods in identifying anticancer peptides.
• To enhance understanding of anticancer peptide mechanisms through feature interpretability.

Main Methods:

• A deep learning model, ACPred-BMF, was developed for anticancer peptide prediction.
• Utilized quantitative and qualitative amino acid properties and binary profile features for numerical representation of peptide sequences.
• Employed a Bidirectional Long Short-Term Memory (LSTM) network architecture with an attention mechanism.
• Applied feature visualization and Shapley Additive Explanations (SHAP) for model interpretability.

Main Results:

• ACPred-BMF demonstrated state-of-the-art performance in predicting anticancer peptides.
• The model effectively utilizes sequence-based features for accurate classification.
• Feature importance analysis provided insights into the characteristics of anticancer peptides.

Conclusions:

• ACPred-BMF offers a reliable and efficient computational approach for identifying anticancer peptides.
• The developed predictor can accelerate the discovery of novel peptide-based cancer therapeutics.
• The interpretability features enhance the understanding of anticancer peptide mechanisms, facilitating drug design.