Article Preview
Top1. Introduction
Knowing the phenotypic drug response on cancer cell is very essential and it serves a major part in determining anti-cancer drug and its prediction. The Genomics of Drug Sensitivity in Cancer (GDSC) dataset offers public access of data to investigators in phenotypic assessment for constructing and testing their techniques. So far, various methodologies have been devised to detect drug sensitivity but they are mostly based on molecular fingerprints or characteristics of drugs rather than their infrastructures. Traditionally, drug discovery is phenotypic and small organic constituents possessing noticeable phenotypic activity are predicted. Among them, an eminent example is penicillin, which is unfortunately determined using phenotypic activity. Phenotypic evaluation, a drug screening architecture has gained much attention among researchers in recent years, because it has found number of approved drugs through this phenotypic screening than that of molecular target-based techniques (Liu, et al., 2019; Szklarczyk, et al., 2019). However, an accurate prediction of drug response in a cancer patient still resembles as an ill-posed problem in oncological medicine. Due to rapid evolution and tremendous growth in data sciences, researchers have preferred computational techniques to understand drug inhibition impacts on cancers depending upon genomics and transcriptomes. Additionally, a general epigenetic modification and DNA methylation has been associated with both development of cancer and drug efficiency. Hence, it is very significant for enhancement of drug response detection by establishing a correlation among drug effectiveness and DNA methylation (Yuan, et al., 2020). Since the drug sensitivity researches continuously producing drug response information, a general question arises if generalization performance of classical detection techniques can further be enhanced with high number of training samples (Xu, et al., 2019; Partin, et al., 2021).
Recently, the study of predicting technique of drug sensitivity has gained more attention because of immense growth in accurate medicine and open accessibility of high pharmogenomics databases (Ledda, et al., 2019; Fusini, et al., 2016). Drug sensitivity detection (Gopal, 2020) is an essential role of customized therapy that is defined as a therapy adaptable to each and every patient and this concept of customized therapy has been already in use since the period of Hippocrates in 5th BC and he cured people depending upon their imbalances of body constituents (Zhang, et al., 2018; Niz, et al., 2016). Usually, it requires a complete study over different drugs, diseases, people and profiling techniques that are bounded by duration, expenses and opportunity of drugs that are analyzed (Catani, et al., 2021; Khan, 2020). Thus, the investigators have been utilizing omics dataset from cancer-obtained cell and detective procedures as an alternate for drug sensitivity prediction (Ahmed, et al., 2020; Chang, et al., 2018; Ganeshan, 2020). Classical machine learning techniques for drug sensitivity detection considers training and testing information that should be similar in feature space and also possess identical underlying representation. In real-time applications, this cannot be taken into an account as certain times one may have enough training sample for processing of drug sensitivity detection in people, whereas the auxiliary information for this specific process are in various feature dimension or may possess various representation. In such circumstances, transfer learning is mostly preferable that would enhance performance of detection models on test data of target process through elevating auxiliary information from related process (Turki, et al., 2017; Gokulkumari, 2020).