Authors:

David Tanase

Date Created:

Not specified

Course Title:

Professor:

Not specified

About Paper:

Colorectal cancer (CRC) has become an increasingly prevalent health issue, especially among younger populations. A particularly challenging subtype, young-onset colorectal cancer (yCRC), is often diagnosed at later stages with metastatic characteristics, increasing treatment difficulty and lowering survival rates. Traditional diagnostic approaches typically focus on single-omics biomarkers, but a more comprehensive approach may lead to higher sensitivity and specificity when it comes to survival prediction, allowing hospitals to allocate resources accordingly. This study proposes an innovative methodology by integrating multi-omics data for predictive performance in the survival prediction of CRC patients. The primary objective of this research is to evaluate whether the integration of multi- omics data can significantly improve predictive machine learning models for patient survival. To train the models, a custom asynchronous downloader was developed to access the TCGA-COAD dataset via the Genomic Data Commons (GDC) API. This allowed for the collection and organization of the multi- omics data that TCGA offers in a format that is best suited for multi-omics analysis. The methodology for model development generally follows two stages. In the first stage, early fusion models are set up for each data type either using pre trained transformers (e.g. ResNet) or custom-built models to generate embeddings. These embeddings are then concatenated into a single unified representation of the patient. The combined embedding is then used as input to a final classification model, trained using a Cox proportional hazards loss function. To evaluate the model, concordance index, partial AIC, and a log- likelihood ratio test are monitored. Poster #2 Histopathology Techniques for Hydrogel-Based Biomaterials William H. Otte

Source:

Texas A&M University / 2025

Topics:

No topics listed

Co-authors:

David Tanase