Development of machine learning models and usage of computational chemistry tools to predict the activity of potential anticancer compounds against K-Ras protein

Abstract

Cancer is a fatal disease with the main identified cause being the alterations in DNA. KRAS is an oncogene highly linked to cancer that appears in the human genome and encodes the respective protein. In the present thesis, the mutation of KRAS G12D was examined as a target for potential anticancer compounds with the use of machine learning and computational chemistry tools. As a first step, a dataset of 530 preprocessed compounds with a total of 208 descriptors (features) was used to train the machine learning models. The significant descriptors that could distinguish the potential anticancer compounds were characterized. Following that, molecular docking experiments were performed between the target protein KRAS G12D and a total of 2000 natural compounds. The experiment was performed using two software programs in an effort to identify compounds that predicted to present inhibitory activity against the examined protein. The results analysis indicated that 6 natural products presented promising results. At last, the important descriptors were calculated for these compounds, and based on them, they were classified. 2 of them were categorized as potential anticancer compounds with satisfactory probabilities of 0,7 and 0,88. Our results were optimistic and can lead to the conclusion that machine learning can be combined with molecular docking to select compounds with anticancer activity