ACPred-LAF

ACPred-LAF predicts anticancer peptides (ACPs) using a multi-sense-scaled attention deep learning architecture to improve identification and prioritization of ACP candidates.

Key Features:

• Multi-Sense-Scaled Attention Architecture: Employs a multisense and multiscaled embedding algorithm with attention mechanisms to learn contextual sequential characteristics of peptide sequences.
• Self-Adaptive Embedding Features: Uses self-adaptive embedding features that dynamically adjust to different datasets and reduce reliance on manual feature engineering.
• Benchmarking Superiority: Demonstrates superior performance over existing state-of-the-art methods on both established and newly constructed datasets through comprehensive benchmarking comparisons.
• Robustness Validation: Validated robustness through data interference experiments, confirming model stability under perturbations.
• New Benchmark Dataset (ACP-Mixed): Introduces ACP-Mixed, an integrated benchmark dataset combining existing datasets to mitigate evaluation bias.

Scientific Applications:

• ACPs identification: Facilitates identification of novel anticancer peptides (ACPs) from peptide sequence data.
• Drug discovery screening: Supports screening and prioritization of ACP candidates in anticancer drug discovery pipelines.
• Computational peptide research: Provides a benchmark dataset and model framework for bioinformatics studies of peptide-based therapeutics and ACP prediction.

Methodology:

A deep learning framework that incorporates multi-sense-scaled attention mechanisms and self-adaptive embedding features to automatically learn complex sequence features without manual feature engineering.