Austin, Minn. — David Guinovart, PhD, and Eric Rahrmann, PhD, assistant professors at The Hormel Institute, University of Minnesota, are the recipients of a two-year, $100,000 Data Science Initiative (DSI) Seed Grant from the University of Minnesota. Their funded project, MOOBI: Multi-Omics Optimization-Based Integration for Enhanced Cancer Research Datasets, aims to tackle key challenges in integrating multi-omics data for more precise breast cancer diagnosis and therapeutic targets.
Multi-omics is a biological analysis approach that makes use of multiple types of datasets. In this project, Dr. Guinovart and team are leveraging data made publicly available from The Cancer Genome Atlas (TCGA). They aim to develop a new, integrated dataset with the ultimate goal of enhancing breast cancer subtype classifications and identifying biomarkers for diagnosis and therapeutic targets. This means patients could have a higher likelihood of being matched with the right treatment options for the right cancer as early as possible.
With his background in applied mathematics, Dr. Guinovart sought Dr. Rahrmann’s expertise in cancer biology and metastasis for this interdisciplinary collaboration.
“The way we see this collaboration is an integrative process: we develop an idea, Eric will offer his ideas, and we will adjust as needed. It keeps the model not only accurate, but also fresh,” Dr. Guinovart said.
“In this particular case, we are trying to add more information to this already available data, developing a model that can respond to real-life problems more effectively. I think this could be used for other opportunities, but we will also have a clean dataset that has been validated at another level. We will also be able to share this ready-to-use data to fit other models, ideas, or research,” Dr. Guinovart said.
With Dr. Guinovart an applied mathematician and Dr. Rahrmann an expert in developmental biology, cancer biology and metastasis, this endeavor is an interdisciplinary collaboration that helps keep the ideas considered and models developed accurate and fresh.
“Too often, we only look at the tip of the iceberg and ignore the rest of the data. Essentially, we’re going back with these new, innovative approaches to revisit old questions and ultimately identify new biomarkers and therapeutic targets,” Dr. Rahrmann said.
The project also holds potential for broader applications in the future. Dr. Rahrmann suggested, for example, that the data may be helpful in identifying transition phases of cancer toward hybrid cancer types at critical times of disease progression.
The TCGA has a treasure trove of data — 2.5 petabytes, in fact, or 2.5 million gigabytes — that has been gathered over decades of research. With so much information at hand, projects like this can find new connections and applications that have yet to be discovered.
Once the research team has its refined, well-structured data, they will feed that data to machine learning models that use multiple parameters for optimized classification to minimize false positives in cancer diagnosis as much as possible.
Dr. Guinovart also commented that The Hormel Institute’s culture encourages innovation and recognizes the value of developing novel methods, which allows for unique opportunities like this one to apply mathematics to real-world applications.
“The HI allows a unique opportunity for this to happen organically. You can learn and see what other scientists are working on, what kinds of questions they are asking,” Dr. Guinovart said.
Post-Doctoral Associate Mohammed Qaraad, PhD, and Senior Scientist Kayum Alam, PhD, are also contributing to the project.
How does machine learning work?
By looking at lots of examples and improving over time.
Imagine you have a big basket filled with apples and oranges, and you want to teach a robot how to tell them apart. At first, the robot doesn’t know what an apple or an orange is, so you start showing it examples.
You pick up an apple and say, “This is an apple,” and then an orange and say, “This is an orange.” The robot starts looking for patterns; maybe oranges are round and bright orange, while apples are sometimes red or green and a little less round.
Once the robot has seen enough examples, you give it a new fruit and ask it to guess. If it identifies a fruit as an apple, but it's actually an orange, you correct it, and the robot learns from the mistake. The more fruits it sees, the better it gets at recognizing them.
Just like teaching a robot to distinguish between apples and oranges by showing it examples, we can train a machine learning model to classify different cancer subtypes using multi-omics data.
Instead of looking at color and shape, the model analyzes patterns in gene expression, mutations, epigenetic markers, and protein levels. By feeding it labeled examples—where we already know the cancer subtype—it learns which features are important for distinguishing them. Then, when given a new patient’s data, the model predicts the most likely subtype.
Just like the fruit example, the more high-quality data the model sees, the better it becomes at making accurate classifications, ultimately helping doctors personalize treatments based on the specific characteristics of a patient’s cancer.
“Ultimately, if we speed that up, we speed up helping people,” said Dr. Rahrmann.
MEDIA CONTACT
Matthew Manguso
Marketing & Communications Manager
[email protected]
ABOUT THE HORMEL INSTITUTE
The Hormel Institute is an independent biomedical research department within the University of Minnesota’s Office of the Vice President for Research. Collaborative research partners include Masonic Cancer Center UMN (a Comprehensive Cancer Center as designated by the National Cancer Institute, NIH), Mayo Clinic, and many other leading research centers worldwide. The Hormel Institute, which tripled in size in 2008 and doubled again in size in 2016, is home to some of the world’s most cutting-edge research technologies and expert scientists. Over the next few years, The Hormel Institute will broaden its impact through innovative, world-class research in its quest to improve human health.