Genetic Disorder and Subclass Prediction based on Machine Learning Approaches

A genetic condition is an ailment resulting entirely or partially from a deviation in the Deoxyribonucleic Acid (DNA) sequence from the standard configuration. Genetic disorders may arise from a mutation in a single gene, mutations in multiple genes, a combination of genetic mutations and environmental influences, or chromosomal damage. It is possible for these mutations to result in lethal diseases. The use of artificial intelligence and its subbranches shown its extreme predicting power in various diseases. The potential of these technologies can be employed to predict genetic problems at an early stage using genomic data for prompt intervention. This research leverages the chi-square test and genetic algorithm for feature selection which enhances the prediction of diseases using machine learning algorithms (Random Forests, Gradient Boosting, XGB, CatBoost, LightGBM). Further, two multi-class classifiers were created: one for predicting disorder categories (mitochondrial, multifactorial, and single-gene) and another for subtypes (nine disorders). The LightGBM classifier achieved the highest accuracy of 92.48% for predicting genetic disorder classes. For subtypes, GB, CatBoost and LightGBM attained a maximum accuracy of 67.08%.