Prediction of drug targets for pancreatic cancer using machine learning techniques

Abstract

Pancreatic cancer is one of the deadliest cancers with a very low survival rate. However, people who are diagnosed early have much longer survival than the ones who are not diagnosed with early screening. Therefore, the importance of early diagnosis and consequently, the treatment of pancreatic cancer can be understood. As pancreatic cancer is rare, early screening for pancreatic cancer is extremely costly. Research has been going on to find such techniques that can detect and hence diagnose pancreatic cancer early through Machine Learning models and use them even for the prediction of survival, Immunotherapy response, risk of re-occurrence, etc. The successful implementation of this technology in the prediction of the presence of pancreatic cancer is a breakthrough as it will greatly increase the survival rate as well as the life expectancy of such patients. One of the major challenges in the treatment of pancreatic cancer is the lack of specific and effective drug targets. In recent years, advances in our understanding of the biology of pancreatic cancer have led to the identification of several potential drug targets, including oncogenic signaling pathways, and cellular metabolism. Pancreatic cancer cells are highly metabolic, relying on glycolysis and the citric acid cycle to generate energy. Inhibiting these metabolic pathways has been shown to reduce the growth and survival of pancreatic cancer cells in preclinical studies. Pan-Cancer dataset from Genomics of Drug Sensitivity in Cancer (GDSC) was used in this research to predict drug targets. In this study Machine Learning algorithms were used such as feature importance using Random Forest, prediction of Drug Targets using Bagging, Dense Neural Network, Naïve Bayes, Multilayer Perceptron, K-Nearest Neighbors, Support Vector Machines, Long Short-Term Memory, Recurrent Neural Network and XGBoost classifier.