Prediction of drug combinations' synergy score by ensemble learning

Recently, the most preferred method in the treatment of complex diseases such as cancer, the origin of which is due to more than one genetic and environmental causes, is polytherapy (combination therapy). It is a method of where more than one drug is used together. If the effect of a drug combination on the cell line with any disease is greater than the sum of the effects achieved by applying the drugs in the combination alone, this drug combination is called a synergistic drug combination. In recent studies in this field, It has been observed that machine learning methods provide efficiency for determining synergistic drug combinations in many aspects such as time, resources, etc. This thesis consists of two parts. In the first part, the effects of data sets that we created with different drug representations on the estimation of synergy scores which show the degree of synergism of drug combinations were examined. Some of the drug representations used for the first time for synergy score estimation. The best results in the literature were obtained for the estimation of synergy scores in a comprehensive oncology dataset by combining machine learning predictions' for these datasets. In the second part, we tried to create second drugs (molecules) for drug-cancer cell line pairs that would optimize synergy scores predicted by an artificial learning model. For this purpose, variational autocoder and gradient ascent methods were used. As a result of this study, it has been observed that, machine learning methods can create molecules that are similar with the molecules that give scores close to the synergy scores that are optimized.