Quantum Machine Learning for Prediction of Compound-Protein Interactions in Drug Discovery

This paper focuses on quantum machine learning (quantum SVM) modeling for the identification of compound-protein interactions (CPIs) crucial in drug discovery and pharmacology. Estimation of various CPIs is one of the prerequisites for selecting potential drugs, improving the activity of lead compounds and studying the toxic effects of drugs. The quantum SVM model introduced above provides higher accuracy and better processing time compared to the quantum computing principles of q-features, q-kernels and classical models. In testing the model, two different datasets were used, a human dataset with 1,052 compounds and 852 proteins from twenty-one different categories and a C. elegans dataset with 1,434 compounds and 2,504 proteins from twenty-one different categories. Quantum SVM is characterized against three traditional machine learning algorithms, among the algorithms of supervised learning, K-Nearest Neighbors (KNN), Random Forest (RF) and Classical Support Vector Machine (SVM) are the most used algorithms. In tuning the decision model, evaluation criteria such as area under the curve (AUC), precision and recall were applied. It can be observed from the results reported in this study that quantum SVM gives better performance when compared to traditional models in terms of performance indices on both datasets. In particular, quantum SVM yielded the highest AUC values (0. 923 for the human dataset and 0. 927 for the C. elegans dataset), meaning a higher ability to rank true interacted compound-protein pairs compared to false positives. Furthermore, quantum SVM has better precision and recall, which can reduce the number of false positives and increase the number of correct interactions required for effective drug candidate identification and cost reduction in experimental validation.