iACP-DRLF

iACP-DRLF predicts anticancer peptides from peptide sequences using deep representation learning features to prioritize candidates for cancer therapeutics research.

Key Features:

• Deep Representation Learning (DRLF): Derives deep representation learning features from peptide sequences to improve discrimination between anticancer and non-anticancer peptides.
• Soft Symmetric Alignment Embedding: Uses alignment-based methods enhanced by neural networks to generate sequence embeddings.
• UniRep Embedding: Employs UniRep deep neural network embeddings based on long short-term memory (LSTM) models to produce comprehensive sequence representations.
• Light Gradient Boosting Machine (LightGBM): Integrates LightGBM as the classifier to distinguish anticancer peptides using DRLF.
• UMAP (Uniform Manifold Approximation and Projection): Applies UMAP for dimensionality reduction and visualization of high-dimensional embedding features.
• SHAP (Shapley Additive Explanations): Uses SHAP to explain feature contributions to model predictions and highlights the superior performance of UniRep embeddings in identifying anticancer peptides.

Scientific Applications:

• High-throughput screening: Rapidly screens peptide sequence datasets to identify potential anticancer peptides in bioinformatics and computational biology workflows.
• Candidate prioritization: Prioritizes peptide candidates for experimental validation to focus wet-lab efforts on the most promising sequences.
• Cancer peptide discovery: Supports discovery and analysis of peptide therapeutics for human cancers by providing interpretable machine-learning-based predictions.

Methodology:

The method trains machine learning models using deep representation learning features from peptide sequences, integrates sequence embeddings (Soft Symmetric Alignment and UniRep) with a LightGBM classifier, and employs UMAP for dimensionality reduction and SHAP for model interpretability.