A Machine Learning Framework for Cancer Prognostics: Integrating Temporal and Immune Gene Dynamics via ARIMA-CNN

Hepatocellular carcinoma (HCC) poses a significant global health challenge due to its high incidence and mortality rates. Our study investigates the prognostic significance of chemokine (C-C motif) ligand 5 (CCL5) and various immune gene signatures in HCC using an innovative combination of Autoregressive Integrated Moving Average (ARIMA) and Convolutional Neural Network (CNN) models. Time series data were utilized to apply an ARIMA model that captures the temporal dynamics of CCL5 expression. This models residual was integrated with immune signature expression data, including lymphocytes and macrophages, to extract features using a CNN model. Our study demonstrates that CNN-extracted features yield a statistically more robust association with patient survival compared to the traditional median split method, which primarily focuses on single-gene analysis. Specifically, CNN-extracted features from CD8 T cells and effector T cells resulted in a hazard ratio (HR) of 0.7324 (p = 0.0008) with log-rank p-value (0.0131), underscoring their pivotal role in the anti-tumor immune response. This methodology highlights the superior prognostic value obtained through integrated multi-gene analyses, providing deeper insights into tumor-immune interactions than conventional single-gene approaches. Moreover, clustering immune genes based on non-parametric correlations unveiled distinct survival patterns. A cluster comprising B cells, Th2 cells, T cells, and NK cells exhibited a moderate protective effect (HR: 0.8714, p = 0.1093) alongside a significant log-rank p-value (0.0233). However, the cluster, including granulocytes, Tregs, macrophages, and myeloid-derived suppressor cells, showed no significant survival association, highlighting the intricate immune regulation within the tumor microenvironment. These findings emphasize the necessity of incorporating temporal dynamics and synergistic immune gene interactions for more accurate prognostic evaluations. Our integrated ARIMA-CNN framework represents a significant advancement, leveraging both linear and nonlinear modeling to uncover the dynamic influence of multiple immune genes. This framework holds excellent potential for identifying robust biomarkers and personalizing immunotherapy strategies, ultimately paving the way for innovative cancer management solutions.