Oluwatosin Oluwadare, Assistant Professor of Computer Science, is seeking to answer a fundamental question in genome science: how the physical structure of a chromosome affects genetic function, from gene expression to DNA replication and incidence of genetic disease. To do so, Oluwadare’s research group created an entirely new way of predicting chromosome structure — one that is faster and more accurate than nearly all current methods.
How did he do it? Together with Van Hovenga, a graduate student in the Applied Mathematics program at UCCS, Oluwadare created an algorithm that learns and re-learns as it computationally predicts chromosome structure, resulting in ever-more refined 3D models.
The novel method is inspired by curriculum learning, a training technique inspired by the way humans learn. Essentially, the algorithm learns to predict the structure of a chromosome just like a person might, and as it “learns,” it performs better and better over time.
Oluwadare and Hovenga named the method Curriculum Based Chromosome Reconstruction (CBCR). In the future, they believe it can be used to understand how chromosome structure might impact disease occurrence on a chromosome- and genome-wide scale.
“The main idea is to break the data into subsets and order them according to some notion of difficulty, which is subject to the input data and its properties,” said Oluwadare and Hovenga. “You then train a model beginning with the easiest subsets and progressively add on harder data as the training progresses. The purpose of doing this is so that the algorithm can transfer knowledge gained from the easier subsets when learning on the more difficult subsets.”
“We found that this technique improves the convergence of the algorithm, decreases the computational burden of training the model, and improves the reconstructive accuracy of its outputs when compared with other existing methods proposed to solve the chromosome 3D structure reconstruction problem.”
In other words, the new method performs better than most existing methods, and represents a new path forward to understanding genomic activity.
Oluwadare and Hovenga hope that the new method inspires other researchers to apply curricula learning to their work. And to support future research, they developed the model as an open-source tool so that researchers working in the computational genomics field might better visualize chromosomes’ spatial organization.
“In the long run,” Oluwadare said, “this project will aid a better insight into the 3D genome topology and support an advanced study to investigate the relationship between the 3D structure conformation and disease occurrence on a chromosome- and genome-wide scale.”
The research was published by the International Journal of Molecular Sciences in April 2021. It is titled “CBCR: A Curriculum Based Strategy For Chromosome Reconstruction.”
Oluwatosin Oluwadare is an Assistant Professor of Computer Science and Innovation within the College of Engineering and Applied Science at UCCS. His research focus areas include bioinformatics and computational biology, machine learning, deep learning and big data analytics. Oluwadare has a keen interest in researching in machine learning and its various applications, including those that can create positive impact on everyday citizens. For instance, he proposed and led the development of a software app called EyeCYou, which uses AI to provide the facial description of a person to the visually impaired. Learn more about Oluwadare’s work on the College of Engineering website.