An algorithm for drug discovery based on deep learning with an example of developing a drug for the treatment of lung cancer
We have designed an algorithm implemented in a software platform for the development of new anti-tumor drugs in the form of small molecules. Several molecules were generated for the treatment of patients with lung cancer as an example. At the initial stage, we identified the targets for the therapy. Firstly, we evaluated the expression profile of the genes most associated with poor clinical outcome in patients with lung cancer using deep learning. Additional patients data were gained by generative adversarial neural networks (GAN) technology. As a result, a set of genes was successfully selected, which expression was associated with poor prognosis. We identified the genes that could distinguish normal tissue from tumor tissue using another deep learning model that was trained to predict normal and tumor tissue based on gene expression. The other genes were considered as targets for targeted lung cancer therapy. After that, a module was developed that predicts the interactions of inhibitors with proteins. For this purpose, the amino acid sequences of proteins were represented in vector form, as well as formulas of chemical compounds interacting with proteins. In addition, a deep learning-based module was developed that predicts the IC50 in experiments on cell lines. Virtual pre-clinical trials with the selected inhibitors were performed to identify relevant cell lines for laboratory experiments. As a result, the study obtained formulas of several molecules with the predicted binding to certain proteins.