Drug Synergy for Cancer using Computational Intelligence Techniques

Abstract

Cancer still remains a major health concern in the world. It is still a challenge for researchers to develop an effective cancer therapy in most types of cancers. Although cancer presents with a very high mortality rate, there are certain treatments available and the number is growing with new breakthroughs every day. A variety of cancer treatments exist, and the type of treatment a patient receives mainly depends upon the type of cancer they present with and its current stage. Various treatments of cancer exist such as radiation therapy, chemotherapy, drug synergy, hormonal therapy and surgery. Some patients need only one kind of treatment, but most of the patients will receive a combination of these. There are many side effects of aforementioned treatments including hair loss, vomiting, nausea, blood disorders, itchiness, constipation, diarrhea, vaginal dryness, enlarged and tender breast. Researchers are trying very hard to reduce the side effects caused by cancer treatment. Now days, more focus is being given by researchers towards targeted therapy such as drug synergism. It mainly targets the cancerous cells. Drug synergism is a branch which finds the optimal combination of two or more drugs on the basis of DDI (Drug Drug Interaction) to kill the cancerous cells in a more efficient manner as compared to their individual effects. However, there are huge number of drugs which leads to the combinatorial explosion problem. To handle the numerous amounts of drugs, machine learning (ML) plays a major role in this area. ML is categorized in supervised and unsupervised learning. In supervised learning, different problems can be categorized under classification and regression whereas in unsupervised learning, clustering problems are considered. Problem of drug synergy falls under the category of regression problems in which synergy score (considered as target) of different drugs is predicted from the given set of input parameters. ML techniques can be used to improve the results of such type of problems. There are some studies exist to predict the synergy score. But these studies have some limitations such as use of single model, a smaller number of input parameters, and lack of drug data to train the models. Because of such issues, the trained model may or may not produce a reliable and efficient prediction. Single model can be replaced with the ensemble models to predict synergy score. Ensemble learning is a process of combining more than one model or combination of one model with some optimization algorithm to solve a given computational intelligence problem. Generally, it is used to enhance the predictability as well as to improve the robustness of a model. The ensemble approach among different models is used because it is capable of boosting the weak learners. This approach has advantage that the ensemble model can adapt any diversity in the data more correctly as compared to single model . Neuro-fuzzy based ensembling technique is designed using biased-weighted aggregation(addition of more weights to model with higher prediction score). Four with the highest accuracy among nine ML models have been selected for developing ensemble-based machine learning model. On the other fold, ensembling of model with optimization algorithm gives the best result by finding the best combinations of parameters to be tuned for the given model. An effective ensemble based machine learning techniques are defined to predict drug synergy, such as SVM based differential evolution and evolutionary based neural network structure using Adaptive Lévy method to overcome the parameter tuning issue. It regulates the precision in prediction. It suggests that the ensemble approach is more efficient than the single model. Therefore, ensemble models have been proposed to predict the drug synergy score. To check the consistency of proposed ensemble model prediction, repeated k-fold cross validation has been performed. The existing machine learning techniques and this evolutionary based technique are tested on drug synergy score data. An extensive analysis shows the better performance of proposed method over existing techniques, in terms of following evaluation parameters such as accuracy, coefficient of correlation and root mean squared error.