ALRIGMR: Adaptive logistic regression via integrating gene mutation and RNA-seq for liver cancer diagnosis

RNA-seq is often used for early accurate diagnosis and related gene screening of liver cancer, significantly improving patients’ survival rates. Popular diagnostic methods based on machine learning often ignore genes with insignificant differential expression in RNA-seq and fail to characterize the overlapping group effect triggered by a few genes participating in multiple biological pathways. This paper aimed to solve the above problems by developing an adaptive logistic regression via integrating gene mutation and RNA-seq (ALRIGMR). A new data integration strategy was proposed to highlight genes with high mutation rates and insignificant differential expression. The local maximal quasi-clique merger (lmQCM) was used for the overlapping grouping, which was proved to be superior to the weighted gene co-expression network analysis (WGCNA). Relying on differential expression and mutational information, a new criterion for evaluating gene significance was proposed. ALRIGMR achieved a diagnosis accuracy of 88.4% on the external validation set, which is 23.0%, 53.8%, 26.9%, 15.3%, 11.5%, 7.6%, 3.8%, and 7.6% higher than that of eight methods. Five insignificant differentially expressed genes, TP53, TTN, MUC16, ABCA13, and RYR2 were screened, which were confirmed to be closely associated with liver cancer.