A Screening Algorithm for Gastric Cancer-Binding Peptides

© 2019, Springer Nature B.V. Gastric cancer-binding peptides (GCBP) are promising diagnostic and therapeutic agents for gastric cancer management. Their utility lies in their ability to facilitate the early detection of gastric cancer, prevent metastasis, and prevent tumor angiogenesis. In order to promote and accelerate the discovery of more GCBP, this study aims to create a machine-learning classification model that can predict if a given sequence can bind with gastric cancer cells. A systematic literature search was conducted to extract peptides that can and cannot bind with gastric cancer cells. Nine descriptor classes were then calculated for each sequence. The resulting dataset was used to create classifiers using five machine-learning algorithms. Rigorous model optimizations were conducted which included descriptor selection and probability threshold tuning. The combination of the topological descriptor T-scales, and logistic regression were found to satisfactorily predict GCBP class. The optimized classification model exhibited satisfactory accuracy with balanced sensitivity and specificity, and excellent precision. The results brought forward provide the foundation for an alternative screening method for GCBPs. This system is expected to positively contribute in the discovery of new GCBPs, thereby potentially enhancing GC disease diagnostics and management