Explainable-AI to Discover Associated Genes for Classifying Hepato-cellular Carcinoma from High-dimensional Data

Abstract

Knowledge-based interpretations are essential for understanding the omic data set because of its nature, such as high dimension and hidden biological information in genes. When analyzing gene expression data with many genes and few samples, the main problem is to separate disease-related information from a vast quantity of redundant data and noise. This paper uses a reliable framework to determine important genes for discovering Hepato-cellular Carcinoma (HCC) from micro-array analysis and eliminating redundant and unnecessary genes through gene selection. Several machine learning models were applied to find significant predictors responsible for HCC. As we know, classification algorithms such as Random Forest, Naive Bayes classifier, or a k-Nearest Neighbor classifier can help us to classify HCC from responsible genes. Random Forests shows 96.53% accuracy with p < 0.00001, which is better than other discussed Machine Learning(ML) approaches. Each gene is not responsible for a particular patient. Since ML approaches are like black boxes and people/practitioners do not rely on them sometimes. Artificial Intelligence(AI) technologies with high optimization interoperability shed light on what is happening inside these systems and aid in the detection of potential problems; including causality, information leakage, model bias, and robustness when determining responsible genes for a specific patient with a high probability score of almost 0.99, from one of the samples mentioned in this study.