GeneXAI: Influential gene identification for breast cancer stages using XAI-based multi-modal framework

Abstract

To provide improved treatment prediction and prognosis, analysis of the categorization of cancer stages and important genes in each stage is necessary. The study introduces a GeneXAI multi-modal approach, which classifies the cancer stages and identifies the influential genes by the explainable artificial intelligence models. In the first phase of the GeneXAI, a hybrid optimal feature selection method is applied to extract the imperative features using an early fusion technique. By using the imperative features, the stages of tumor, lymph nodes, and metastasis are identified, and finally, the accurate stage of cancer is classified. In the second phase, XAI such as SHAP and LIME has been utilized to identify the best genes for distinct cancer stages. Moreover, the genomic dataset's top genes were found using SHAP, while crucial genes were found by instance using LIME. Some influential genes such as (PLA2G10, MST1R F13B, and CAMK1) identified by the GeneXAI model, have also been recognized as equally important genes in the state-of-the-art biological models. The model illustrates the process of classifying the influential genes as prognostic or non-prognostic based on their clinical importance. The proposed framework achieves an average 5–7% higher accuracy than other state-of-the-art models by using the early fusion technique of a multi-modal approach.