Performance Evaluation of Quantum Machine Learning Classifiers in Bioinformatics Applications

Abstract

Bioinformatics is a field that requires Artificial Intelligence (AI) capabilities to analyze and process large datasets. One solution currently reported to handle the problems of high complexity in AI models is Quantum Machine Learning (QML). QML is a way to process large datasets more quickly and accurately by using the rules of quantum physics. This research evaluates the ability of two QML methods, Quantum Support Vector Classifier (QSVC) and Variational Quantum Circuit (VQC), to predict off-target effects in CRISPR/Cas9. The dataset used has a very high data imbalance, making it a challenge for the model to handle the imbalance. To achieve the best model performance, our research uses hyperparameter optimization to determine the optimal parameters for the number of qubits and circuit depth with linear entanglement and ZZFeatureMap. Our evaluation performance uses AUROC and PRAUC as the primary references, followed by accuracy and F1-score. The comparison results indicated that QSVC achieved the best performance with six qubits and two repetitions, having AUROC 0.861, PRAUC 0.821, accuracy 0.785, and F1-score 0.781. In comparison, VQC had AUROC 0.713, PRAUC 0.602, accuracy 0.725, and an F1-score of 0.727. The results indicate that QSVC with appropriate parameters can provide significant results for imbalanced data in the case study of predicting off-target effects of CRISPR/Cas9. Future research is expected to process more datasets with QML enhancement to handle a larger number of datasets using more optimal resource utilization, addressing the analytical needs in the field of bioinformatics.

Access to Document

Other files and links

• Link to publication in Scopus

Fingerprint