Publication Date
Fall 2025
Degree Type
Thesis
Degree Name
Master of Science (MS)
Department
Computer Engineering
Advisor
Carlos Rojas; Bernardo Flores; Mahima Suresh
Abstract
The study of genomic data is important for interpreting their function. Hi-C (high-throughput chromatin conformation capture) data gives researchers a way to analyze the 3-D structure of a genome. This helps researchers understand how distal interactions activate genes. While Hi-C data offers important insights into the mechanisms underlying diseases such as cancer, its limited resolution constrains a comprehensive view of interactions. This study investigates the potential of generative models to enhance the resolution of Hi-C maps, with a emphasis on leveraging the close relationship between text- and image-based embedding models. This work focuses on modifying existing architectures to improve results. Embedding the transformer architecture into Akita yielded notable performance gains. Another set of experiments focused on using a modified version of the Muse architecture and this showed promise. These findings suggest that generative models hold strong potential for working effectively on Hi-C data, and this work serves as an initial step toward advancing research in this direction.
Recommended Citation
Choksi, Manan, "Advancing Hi-C Resolution Through Generative And Transformer-Based Models" (2025). Master's Theses. 5736.
DOI: https://doi.org/10.31979/etd.jhxm-55w7
https://scholarworks.sjsu.edu/etd_theses/5736
