Means of Regression Methods for Modelling and Prediction of Drug Score Based on Cell Lines Data

Annually more than 10 million patients worldwide are diagnosed with cancer. Meanwhile, according to the statistics, cancer is one of the main causes of death in the world: mortality reaches 20%. Nowadays, there is no universal treatment for this disease. Consequently, the choice of the appropriate treatment for each concrete patient is an important step, because otherwise the treatment may give no effect, while the cost of treatment in many cases is high. In addition, the time factor is very important in cancer therapy. And what is the worst, the wrong medicine might cause additional damage to the patient. Over the past few decades, main success in oncology has been achieved mainly due to the emergence of a large group of new types of drugs aimed at suppressing the vital properties of cancer cells. In addition, it turned out that for two different patients with the same type of cancer, these drugs act differently, and this posed the problem of individual selection of the drug for each particular patient. This work is devoted to the analysis of different regression models and their ability to predict the effectiveness of treatment with a specific drug. For this purpose, in the course of experiments the definite procedure is used for the selection of regression models, which are subsequently used to predict the result of patient treatment and to assess the accuracy of classification of patients to «responders» and «non-responders». As the result of the experiments, a number of conclusions are made about how different regression models work, how different types of regularization influence on accuracy of classification models. In addition, for each dataset there are certain patients’ characteristics that play the most important role in the treatment outcome prediction. Finally, for most of the data sets, constructed models predict the effectiveness of treatment with an accuracy exceeding the results of previous work.