A Critical Evaluation of Background Gene Omission in Imaging Transcriptomics

Imaging transcriptomics integrates spatial gene expression data with imaging-derived phenotypes (IDPs) to elucidate molecular mechanisms that underlie brain structure and function. Overrepresentation analysis (ORA) is widely used to annotate IDP-related genes; however, many studies have overlooked appropriate background gene selection. Here, we critically evaluated the impact of omitting a proper background on ORA findings. A systematic review of 152 imaging transcriptomics studies (2015–2024) revealed that 84.9% did not report background genes, and only 5.26% used the Allen Human Brain Atlas (AHBA) genes as background. Simulations showed that ORA significance increased with background size. In realistic simulations, default backgrounds (e.g., all protein-coding genes) inflated pathway significance by up to 50-fold, with probabilities reaching 0.97, particularly for frequently reported pathways related to synaptic signaling and neurotransmission. In contrast, using AHBA genes as the background maintained the significance probabilities near 0.05. These findings highlight the need for appropriate background selection and transparent reporting and we provide practical guidance for ORA in imaging transcriptomics. Overrepresentation analysis (ORA) is widely used in imaging transcriptomics to identify biological pathways linked to brain imaging findings. However, many studies rely on the default options provided by ORA tools, which often use improper background gene sets and can lead to inflated or misleading results. In this study, we reviewed the literature and conducted simulations to show that improper background selection (e.g., using all protein-coding genes) can bias pathway enrichment. We highlight the importance of using appropriate background genes and provide recommendations for best practices in ORA for future studies.