A Comprehensive Review of Various Machine Learning and Deep Learning Models for Anti-Cancer Drug Response Prediction: Comparative Analysis With Existing State of the Art Methods

Abstract

The optimal treatment selection for cancer patients is extensive, and pharmacogenetic prediction is done using genetic cohort, chemical structure, and target information. Though previous studies sought to characterise pharmacological reactions, there were limits in categorization. Due to the development of various solutions, existing feature selection techniques such as statistical combinations suffer from drawbacks such as local optima, lack of heuristics, and so on. This further leads to a low convergence rate which affects the classification rate. To address this, the current study describes a hybrid approach that is based on machine learning and deep learning, as well as a comparison of the localization heuristic-based Harris Hawk intelligence method and Gravitational Optimization methods with Machine Learning (ML) and Deep- Learning (DL) algorithms. The study suggests the use of Conditional Generative Adversarial Network (CGAN) to obtain better feature selection with less volatility in order to improve data quality and minimise intrinsic variation. In this study, the possible associations between cell lines and drugs are deduced using the CCLE- Cancer-Cell Line Encyclopaedia and Genomics of Drug Sensitivity in Cancer- GDSC datasets, and the study proposes a hybrid Bi-Residual Dense Attention Network for cell line categorization. The proposed method shows better prediction performance based on precision, accuracy, F1-score, Area under curve (AUC), Area under the receiver operating characteristic curve (AUROC), specificity and recall. For the GDSC dataset, the BRDAN-HH framework achieved an accuracy of 0.9675, recall of 0.9795, specificity of 0.975, precision of 0.9785, F1-score of 0.9799, AUC of 0.97, and AUROC of 0.9705. Similarly, for the CCLE dataset, it demonstrated robust performance with an accuracy of 0.9655, recall of 0.986094, specificity of 0.975, precision of 0.975, F1-score of 0.986, AUC of 0.966, and AUROC of 0.9758. The results highlight the efficacy of the BRDAN-HH framework in delivering superior classification metrics, making it a valuable tool for analysing large-scale biomedical datasets.