mACPpred 2.0

mACPpred 2.0 predicts anticancer peptides (ACPs) from primary amino acid sequences using stacked deep learning with integrated spatial and probabilistic feature representations.

Key Features:

• Stacked Deep Learning Architecture: Employs 1D convolutional neural network (1D CNN) blocks combined with hybrid features to identify hidden patterns and improve ACP prediction accuracy.
• Integrated Feature Representations: Combines spatial features from the top seven performing NLP-based embeddings with 90 probabilistic features, representing the first integration of spatial and probabilistic feature representations for ACP prediction.
• Comprehensive Feature Evaluation: Evaluates both conventional feature descriptors and advanced pre-trained natural language processing (NLP)-based embeddings using eleven different classifiers.
• Updated Benchmarking Dataset: Constructs an up-to-date dataset by integrating all publicly available ACP datasets to reflect the exponential growth in characterized ACPs.
• Rigorous Validation: Performance assessed through cross-validation and independent tests demonstrating superiority over mACPpred and existing state-of-the-art predictors.

Scientific Applications:

• Anticancer Peptide Identification: Enables researchers to identify potential ACPs from primary amino acid sequences for further experimental testing and validation.
• Immunoinformatics Research: Addresses the major challenge of ACP identification based solely on sequence information, advancing computational approaches in cancer immunotherapy.
• Comparative Predictor Evaluation: Provides a benchmark for comparing performance against existing machine learning-based ACP prediction methods.

Methodology:

The method employs a stacked deep learning approach incorporating 1D convolutional neural network blocks with hybrid features combining the top seven performing NLP-based embeddings and 90 probabilistic features. Eleven different classifiers were evaluated on conventional feature descriptors and advanced pre-trained NLP-based embeddings using an integrated benchmarking dataset of all publicly available ACP datasets.