ACPred-BMF

ACPred-BMF predicts anticancer peptides (ACPs) from peptide sequences using deep learning and interpretable feature representations to support identification of candidate therapeutic ACPs.

Key Features:

• Deep learning-based prediction: Employs a deep learning predictor to classify peptide sequences as anticancer peptides.
• Quantitative and qualitative amino acid properties: Utilizes both quantitative and qualitative properties of amino acids in peptide sequences for feature representation.
• Binary profile feature transformation: Transforms peptide sequences into numerical representations using a binary profile feature approach for model input.
• Bidirectional LSTM (BiLSTM) network: Uses a Bidirectional Long Short-Term Memory network architecture to capture dependencies in peptide sequences.
• Attention mechanism: Integrates an attention mechanism to weight relevant sequence positions during prediction.
• Feature visualization techniques: Incorporates visualization methods to investigate model-learned features and sequence patterns.
• SHAP (Shapley additive explanations): Applies SHAP to determine feature importance and provide interpretable explanations of predictions.
• State-of-the-art performance: Reported empirical results indicate leading predictive performance among ACP predictors.

Scientific Applications:

• High-throughput screening: Enables high-throughput screening of peptide sequences to identify candidate anticancer peptides.
• Mechanistic interpretation: Provides feature importance and visualization to aid understanding of sequence determinants and mechanisms of anticancer activity.
• Support for research domains: Supports research in oncology, medicinal chemistry, and bioinformatics by prioritizing and characterizing ACP candidates.

Methodology:

Computational methods explicitly include binary profile feature transformation of peptide sequences, extraction of quantitative and qualitative amino acid properties, prediction using a Bidirectional LSTM (BiLSTM) with an attention mechanism, and SHAP for feature importance and visualization.