Modelos de Aprendizagem Computacional para previsão da eficácia de medicamentos avaliados em linhas celulares

Abstract

There is currently a significant increase of diseases in the world and, therefore, there is a greater amount of medication available to provide greater effectiveness in health systems. The current availability of medicines at the commercial level is quite relevant, and their benefits and therapeutic importance are irrefutable. Thus, the importance of new medicines is undeniable for human life. There are several stages during the drug design process until it reaches the final phase that corresponds to clinical tests. One step in the drug design process is to test the newly synthesized drug in vivo. This phase is characterized by being an expensive phase in terms of costs as well as painful for the animals that are submitted to the tests. To avoid or at least reduce, animal testing, pharmaceutical companies, and research laboratories opt for a procedure called biopsy where they obtain a small sample of malignant tissue (human or animal), which is quite harmless to the biological supplier. These samples can replicate "ad infinitum" cells in the laboratory where each cell replicated in this way is called a cell line. That way, they can test new drugs in thousands of different doses on samples of these replicated tissues in a less expensive way and monitor their effects as well as their effectiveness. These replicated tissues also allow researchers to test new approaches and have an appropriate process for comparing results. The objective of this dissertation is the application of Data Mining to public data, already used in an original study for the regression problem, that reports the results of the effectiveness of drugs using cell lines. In addition to the algorithms used in the original study, different algorithms were introduced, a feature selection was carried out, as well as the transformation of the regression problem into classification one and respective analysis.