Knowledge Graphs

Understanding the Tumor Microenvironment in Colorectal Cancer: A Knowledge Graph Approach 

The bottom figures (shown avove)  provide detailed information on specific gene-cell relationships within the TME of colon cancer. Each sub-figure focuses on a particular cell type and its associated genes. For instance, the first sub-figure shows the relationship between the enteroendocrine cell of the colon and genes like AGTR1, ANK2, and ABCC8. Similarly, other sub-figures highlight the relationships between tuft cells, stem cells, absorptive cells, progenitor cells, paneth cells, and their respective genes. These detailed mappings are essential for identifying potential biomarkers and therapeutic targets specific to different cell types within the TME, thereby aiding in the development of more precise and effective treatments for colon cancer.  

For example, the predicted clustering of tuft cells (AFAP1L2  marker) and stem cells (ALDH1B1 marker), captured expression here on H&E image (xenium) with their corresponding marker genes and in the knowledge graph suggests a close interaction or shared pathways within the Tumor Microenvironment (TME) of colorectal cancer (CRC). This clustering can indicate important biological events such as cellular differentiation, immune response, or tumor progression. For example, tuft cells are known to play a role in sensing the environment and initiating immune responses, while stem cells are crucial for tissue regeneration and repair. Their close proximity in the graph may reflect their collaborative roles in maintaining the TME and responding to tumor-related changes.

Understanding these clusters can provide insights into the disease biology of CRC and help design targeted immunotherapy strategies. For instance, targeting the interactions between tuft cells and stem cells could potentially disrupt tumor growth and improve therapeutic outcomes. Additionally, integrating spatial omics data like Xenium can further enhance our understanding of these interactions and their spatial context within the TME, providing a more comprehensive view of the tumor landscape and aiding in the development of more effective treatments.

Mapping Mid-Lobular Gene Networks: Unraveling Liver Fibrosis Pathways with GenAI and Multimodal Data Integration 

The knowledge graph illustrates the intricate relationships among genes associated with mid-lobular liver zonation, utilizing a combination of data from histopathology spatial transcriptomics (visium), spatial omics (CosMx), and single-cell RNA sequencing (scRNA-seq). Liver zonation refers to the functional specialization within different zones of liver lobules, influenced by gradients of oxygen, nutrients, and other signals. This zonation is essential for understanding how specific cellular responses vary across these regions under normal and disease conditions, particularly in the context of liver fibrosis.

In this network, genes such as VWF, TIE1, KDR, and APOC1 represent critical factors that play roles in endothelial function, immune response, and metabolism within liver lobules. The interconnections among these genes suggest a complex network of gene-gene interactions and co-expression patterns, highlighting how expression changes dynamically across liver zones in response to damage and fibrotic processes. These relationships are especially relevant because liver fibrosis often initiates in mid-lobular regions, where cells are subject to unique environmental conditions. Identifying these patterns provides insight into potential biomarkers of fibrosis severity and progression, as well as a foundation for targeted interventions.

The knowledge graph emphasizes pathways that could serve as therapeutic targets for fibrosis. Genes involved in angiogenesis (such as TIE1 and KDR) and immune response genes (such as C8A and C9) are particularly relevant. By targeting specific pathways in these networks, it may be possible to modulate the fibrotic process more effectively. Drugs designed to influence these pathways, or key regulatory genes could potentially slow or reverse fibrosis.

The integration of multimodal data from histopathology, spatial omics, and scRNA-seq allows for a comprehensive view of gene expression changes in distinct liver zones and cell types. This layered approach provides a detailed understanding of how fibrosis develops on a cellular level, thus aiding the identification of specific cell types or spatial regions within the liver that are most responsive to fibrotic stimuli.

GenAI (generative AI) plays a crucial role in analyzing this multimodal data, allowing for pattern recognition across vast datasets, hypothesis generation, and the suggestion of drug targets based on detailed molecular profiles. The application of GenAI accelerates the research process by automating data interpretation and identifying potential therapeutic targets, ultimately streamlining drug discovery for liver fibrosis and similar complex diseases. This approach not only facilitates a deeper understanding of disease mechanisms but also expedites the development of targeted treatments tailored to the specific pathology of liver fibrosis. 

Leveraging Graph Neural Networks for Predicting Cell Types in Colorectal Cancer: A Spatial Transcriptomics Approach 

The image presents below a study on modeling colorectal cancer (CRC) using a Graph Neural Network (GNN) to predict cell types, leveraging the Squidpy Visium CRC dataset. The top row shows histological images of CRC tissue sections labeled as "spatial1" and "spatial2," with different regions highlighted. Below these images are network graphs representing cell type interactions within these spatial contexts. The cell types include CK low HR low tumor cells, apoptotic tumor cells, macrophages, T cells, proliferative tumor cells, and small elongated stromal cells. The bottom section of the image shows additional histological images with different cell type annotations labeled as "ID1" and "SPP1."

Benefits of Using This Approach for Targeted Immunotherapy:

1. Spatial Context: GNNs can capture the spatial relationships between different cell types within the tumor microenvironment, providing a more comprehensive understanding of cell interactions.

2. Predictive Accuracy: By modeling the complex interactions between cells, GNNs can improve the accuracy of cell type predictions, which is crucial for identifying specific targets for immunotherapy.

3. Personalized Treatment: This approach can help in identifying unique cellular compositions and interactions in individual patients, leading to more personalized and effective immunotherapy strategies.

4. Data Integration: Squidpy Visium CRC data integrates spatial transcriptomics with histological data, allowing for a multi-dimensional analysis that enhances the understanding of tumor biology and the immune landscape.

This approach demonstrates the application of advanced computational techniques like GNNs in cancer research, highlighting the potential for improved diagnostic and therapeutic strategies in colorectal cancer.

Modeling Cell Types Using Graph Neural Networks and Spatial Transcriptomics 

The top row of the image shows tissue from a mouse brain, as indicated by the Squidpy Visium Fluorescent dataset. The bottom row features network graphs modeling cell types using Graph Neural Networks (GNN) at different neighbor variables (n_neighs=40, 20, 10).

Correlation with Mouse Brain Cell Types:

• Possible Cell Types in Mouse Brain:

  • T Cells: Present in the brain, especially in the context of neuroinflammation.

  • Macrophages: Microglia are the resident macrophages of the brain.

  • Endothelial Cells: Form the blood-brain barrier and are present in the brain vasculature.

• Less Likely Cell Types in Mouse Brain:

  • Tumor Cells: Typically associated with cancerous tissues, not normal brain tissue.

  • Stromal Cells: More common in connective tissues and tumors, not typically found in the brain.

While some cell types like T cells, macrophages (microglia), and endothelial cells are indeed present in the mouse brain, others like tumor cells and stromal cells are more commonly associated with tumor environments. Therefore, the presence of these cell types in the bottom row suggests a pathological condition, such as a brain tumor, rather than normal brain tissue.