GCNCPR-ACPs: a novel graph convolution network method for ACPs prediction.

Area of Science:

• Biochemistry
• Computational Biology
• Drug Discovery

Background:

• Anticancer peptides (ACPs) are crucial for inhibiting and killing tumor cells.
• Identifying ACPs is vital for developing new anticancer therapeutics.
• Predicting ACPs and non-ACPs is a significant area of research.

Purpose of the Study:

• To develop a novel, accurate machine learning-based method for predicting anticancer peptides.
• To improve the identification of potential anticancer peptide drug candidates.

Main Methods:

• Proposed GCNCPR-ACPs (Graph Convolutional Neural Network Method based on collapse pooling and residual network to predict ACPs).
• Utilized residual graph convolutional networks and differentiable graph pooling.
• Extracted features using peptide sequence information, node2vec, and one-hot embedding for amino acid representation learning.

Main Results:

• GCNCPR-ACPs demonstrated superior performance compared to state-of-the-art methods in both ten-fold cross-validation and independent validation.
• Achieved high Matthews Correlation Coefficient (MCC) and Area Under the Curve (AUC) scores.
• Outperformed existing predictors in MCC (69.5%) and AUC (90%) during cross-validation, and MCC (69.6%) and Specificity (SP) (93.9%) in independent testing.

Conclusions:

• The GCNCPR-ACPs method effectively predicts anticancer peptides.
• The proposed method offers significant improvements over existing ACP prediction tools.
• This tool can accelerate the discovery and development of novel anticancer peptide-based drugs.