Discovery of anticancer peptides from natural and generated sequences using deep learning.

Area of Science:

• Biochemistry
• Computational Biology
• Oncology

Background:

• Anticancer peptides (ACPs) offer targeted cancer cell killing, crucial for clinical applications.
• Existing artificial intelligence (AI) predictive models for ACPs often lack robust experimental validation, hindering novel discoveries.
• There is a need for validated AI tools to accelerate the identification and development of effective ACPs.

Purpose of the Study:

• To develop and validate CNBT-ACPred, a novel deep learning model for predicting anticancer peptides.
• To enhance the discovery pipeline for novel ACPs through AI and experimental validation.
• To identify and characterize potent ACP candidates with potential therapeutic applications.

Main Methods:

• Development of CNBT-ACPred, a three-channel deep learning architecture for ACP prediction.
• Extensive in vitro and in vivo experimental validation of predicted ACP candidates.
• Large-scale sequence screening using Uniprot and deep generative models.

Main Results:

• CNBT-ACPred achieved high accuracy (0.9554) and MCC (0.8602), outperforming existing models.
• Identified 37 out of 41 candidate peptides with in vitro tumor inhibitory activity from millions of sequences.
• tPep14 demonstrated significant in vivo anticancer efficacy in mouse models with no detectable toxicity.

Conclusions:

• CNBT-ACPred represents a validated and effective tool for accelerating ACP discovery.
• The study identified promising ACP candidates, including tPep14, for further preclinical and clinical development.
• Established correlations between amino acid composition, structure, and function of identified ACPs provide insights for future drug design.