Uma abordagem de classificação de subtipos de leucemia para identificação de genes diferencialmente expressos utilizando dados genéticos aliado à técnicas de aprendizado de máquina

Abstract

Leukemia is one of the main specific cancer diseases that confer mortality and morbidity in different age groups. The diagnostic challenge is caused by several factors, with the incorrect classification of disease subtypes being one of the main ones. Therefore, it becomes essential to discover the genetic disorders that occurred that caused a certain disease. In this context, the use of machine learning can be applied to solve problems related to leukemia. Therefore, we aim to create a tool for applying a machine learning model that is interpretable and capable of identifying differentially expressed genes and classifying subtypes of leukemia. To this end, a pipeline based on the CRISP-DM methodology was proposed, with the purpose of preparing genetic data and training multi-class classified models. Therefore, different approaches and classifiers were used to determine optimized machine learning models with high accuracy. The models used have canonical and hierarchical approaches, in addition to using feature selection techniques for their training. Highly accurate results were obtained in experiments carried out in comparison with results obtained in the literature, making it possible to compare different approaches, techniques and specific resources. Finally, an application was created addressing the approved concepts for creating machine learning models in an intuitive way and given the need to interpret the results, the SHAP library was used to establish the main genes for global classification and how contributions of each gene to the classification of a given sample.