Combining of Genetic Algorithm and Multiple Linear Regression in Breast Cancer’s Drug Design

Abstract

Breast cancer is the first cause of death by cancer in women. Even so, men could have breast cancer. In the treatment of breast cancer there are surgery, radiation therapy and systemic therapy which treatments using drugs. WHO has listed thirty cytotoxic and anticancer drugs to prevent and reduce breast cancer risk. Researchers have been trying to find other drugs to help people with breast cancer. Thus, drug design becomes more important in discovering new potential drugs to treat breast cancer. In this study, we proposed multiple linear regression (MLR) approach using quantitative structure activity relationship (QSAR) method for modelling drug design of breast cancer. Because the data are obtained from public protein bank have lower number of compounds than the number of features, it failed the assumptions of MLR analysis and led to multicollinearity. QSAR model appeared uncertain when multicollinearity arise. We implemented genetic algorithm (GA) to resolve multicollinearity. GA acted as a feature selector to obtain the most significant features and helped getting the most fitted QSAR model. The experimental result shows that combining of GA and MLR can be implemented in breast cancer's drug design with r-sq \gt 0.38.