Colorectal cancer (CRC) is one of the leading malignant diseases in the United States, predominantly due to its poor prognosis and high metastasis. Tumor-associated macrophages (TAMs) are amongst the most common cells that play a significant role in cancer survival and progression in the tumor microenvironment. By using single-cell CRC-specific RNAseq datasets and computational approaches developed in-house, I aim to answer two specific scientific questions: (Q1) Do TAMs show distinctive signature in CRC samples in contrast with healthy samples?; and (Q2) Can I relate the pro-inflammatory and anti-inflammatory polarization of TAMs to the prognosis of colorectal cancer? I filter macrophage cells from eight publicly-available single-cell CRC-specific RNAseq datasets, obtained from both human (Homo Sapiens) and mouse (Mus Musculus) samples and refine a computational model, called SMaRT, to identify a distinctive signature for accurately predicting macrophage-polarization states in the specialized context of CRC. The computational analysis suggests: (a) TAMs are consistently more reactive in tumorous cells as compared to the healthy cells and that the separation between their source samples is statistically significant; (b) A TAM-specific composite gene signature can be reliably used to separate samples that are cancerous versus samples that are healthy. Specifically, these findings provide sufficient and statistically significance evidence that TAMs have a distinctively different signature in CRC samples, majorly falling in the spectrum of the immuno-reactive polarization state.
In the past two decades, there has been an exponential growth in methods for biomarker discovery for diseases. Despite all the new technology and data available, there is a disconnect between the biomarker discovery phase and application in a clinical setting. Most methods have not provided actionable biomarkers that allow us to design tests for early diagnosis, pass new therapeutics or prognosticate the risk of disease progression. Our lab leverages the abundance of publicly available data and uses the power of Boolean implication analysis to identify and validate biomarkers in multiple independent datasets. In my dissertation, I aim to showcase that using the Boolean implication approach in the context of diseases can provide more information about disease progression compared to traditional methods. In Chapter 1, I use Boolean implication network analysis to identify a gene signature for gastric cancer. I extensively validate this signature in multiple independent datasets and show that this signature can prognosticate the risk of progression to gastric cancer. In Chapter 2, I explore how a Boolean implication analysis identified gene – CDX2 – can be used as a biomarker for epithelial damage in inflammatory bowel disease. I perform a newly developed protocol for immunohistochemistry staining that shows a loss of CDX2 in certain regions of moderate to severe IBD samples. I also demonstrate how CDX2 expression can predict response to various therapeutics. Finally in Chapter 3, I use Boolean implication network-derived macrophage and fibroblast polarization signatures to identify changes in rheumatoid arthritis (RA). I derive RA-specific macrophage and fibroblast signatures that can accurately predict RA samples. Through this dissertation, I demonstrate that the Boolean approach provides a more comprehensive framework to identify more reliable targets for disease diagnosis, therapeutics and prognostication.
Morphological study in system biology provides a broader perspective of understanding biological systems’ structure, form, and organization. Nowadays, incorporating state-of-the-art novel vision-AI techniques revolutionizes this study and could accelerate the feature extraction process and lead to groundbreaking discoveries. The design of novel computer vision-based Deep Learning algorithms enables the development of predictive models, which helps in studying disease progression, developing personalized medicines, drug testing, organ replacement, etc. This thesis presents novel procedures and techniques to extract features from confocal and histopathological images to study organoid culture and colorectal cancer. I have successfully created a unique dataset of Crohn’s disease patient-derived organoids (PDOs) and normal colon tissue samples from mice and humans. Organoids need rigorous rapid imaging for continuous monitoring over a long period. Therefore, it is challenging for scientists to process and verify the data manually. Our developed first-of-its-kind novel organoid mining engine process provides a real-time investigation of organoids. The developed model accurately locates, quantifies, tracks, and classifies human colon organoids without expert intervention. Histopathology image analysis is the key to diagnosing colon cancer by focusing on cell morphology and tissue structures. A pathologist takes images from the interest section of the tissue and prepares them for further analysis. The traditional method involves hand-crafted feature extraction followed by classical image processing techniques. I have introduced an original U-shaped crypt segmentation model using novel vision-AI on colon tissue, revealing a new gene expression pattern on the glandular epithelium cells.
Cookie SettingseScholarship uses cookies to ensure you have the best experience on our website. You can manage which cookies you want us to use.Our Privacy Statement includes more details on the cookies we use and how we protect your privacy.
