Combining Gene Data and Tissue Images to Better Identify Tissue Regions

What is it about?

Understanding how different regions of biological tissues are organized is essential for studying human health and disease. Spatial transcriptomics is a powerful technology that allows researchers to examine gene activity while preserving its location within a tissue. By combining this information with microscopic tissue images, scientists can gain a more complete understanding of tissue structure. However, integrating these two types of information is challenging. Existing methods may rely too heavily on one source while overlooking valuable information from the other. They may also struggle to accurately distinguish different tissue regions. To address these challenges, we developed stMRPC, an artificial intelligence framework that combines gene expression data and tissue images more effectively. Our method automatically adjusts the contribution of each information source and examines tissue patterns from two complementary perspectives, helping identify meaningful tissue regions more accurately. We evaluated stMRPC using datasets from the human brain, breast cancer tissues, and mouse brain. The results demonstrate improved accuracy and robustness compared with several existing methods. Our research provides a useful computational tool for exploring tissue organization and may support future studies of biological processes and diseases.

Featured Image

Photo by National Cancer Institute on Unsplash

Photo by National Cancer Institute on Unsplash

Why is it important?

Accurately identifying different regions within biological tissues is essential for understanding how cells are organized, how they function, and how diseases develop. As spatial transcriptomics technologies generate increasingly rich molecular and imaging data, effectively combining these different sources of information has become an important challenge. Our research addresses two key limitations of existing approaches. First, different data sources do not always contribute equally, which can cause valuable biological information to be overlooked. Second, conventional methods often analyze tissue patterns from a single perspective, potentially missing important structural relationships. What makes stMRPC unique is its ability to automatically balance information from gene expression data and tissue images while examining tissue organization from two complementary clustering perspectives. This allows the model to capture both differences between tissue regions and more subtle relationships within them. Experiments on human brain, breast cancer, and mouse brain datasets demonstrate that stMRPC improves spatial domain identification compared with several existing methods. Additional biological analyses show that the identified regions exhibit meaningful gene expression patterns and functional characteristics. By providing a more accurate and biologically interpretable way to analyze complex tissues, stMRPC could help researchers better investigate tissue organization, explore disease-related changes, and gain deeper insights into the biological mechanisms underlying health and disease.

Perspectives