Integrative Analysis of BMI and Gene Expression Reveals Molecular Interactions Underlying Cancer Progression

Background: Obesity is a chronic condition linked to health issues such as diabetes, heart disease, and increased cancer risk. High body mass index (BMI) is associated with cancers such as breast and colorectal cancer due to hormone imbalances and inflammation from excess fat, whereas a low BMI can raise cancer risk by weakening the immune system. Maintaining a normal BMI improves cancer treatment outcomes, but in some cases, higher BMI might offer protective effects—a phenomenon known as the “obesity paradox”. This study explores how BMI affects gene expression in cancer, using data from The Cancer Genome Atlas (TCGA), aiming to uncover links between BMI and cancer progression while identifying potential treatment targets. Methods: To analyze the data, a two-stage method using overlapping group screening (OGS) was applied. First, gene groups were identified with the “grpregOverlap” R package. Then, their interactions were tested using the sequence kernel association test. Significant gene-gene interactions were selected based on statistical measures. In the second stage, predictive models were built using regularized regression techniques such as ridge regression, lasso, and adaptive lasso, with generalized ridge regression used to improve accuracy and stability in handling high-dimensional data. Results: The proposed OGS-based method was tested on simulated and real-world datasets. Results showed that combining OGS with generalized ridge regression and adaptive lasso (OGSG. ridgeALasso) gave the best prediction performance, with lower error rates and greater stability compared to other models like support vector regression, k-nearest neighbors, and random forests. In practical applications, gene expression and BMI data from TCGA patients (including bladder, cervical, esophageal and liver cancers) were integrated to identify key genes and interactions related to BMI. Conclusions: Through evaluations on both simulated synthetic datasets and real-world datasets, we demonstrated the effectiveness of the proposed method in terms of predictive accuracy. Additionally, we identified BMI-associated genes and gene-gene interaction biomarkers across different cancer types and presented the corresponding network structures. Based on the key genes and gene interactions identified, we further explored how BMI influences cancer development and prognosis, providing deeper insights into the biological mechanisms underlying these associations.