Improving the Rigor of AI and Other Analytical Tools for Genomic Analysis
Researchers are developing new and innovative ways to apply AI and other computational tools to advance research in a wide variety of fields. But how are these tools vetted, and do they consistently work as intended? Graduate student Maxwell Konnaris has been working to answer these questions and make these tools more rigorous and standardized, particularly in the field of genomics.
Researchers regularly develop new methods to analyze genetic and other health-related information, including for the genome, transcriptome, and proteome—the DNA, RNA, and proteins inside an organism—as well as the microbiome—the microbes that live on or inside an organism. These methods may be used, for example, to identify patterns in the sequence of letters in the DNA alphabet, how strands of DNA are packaged in the cell, or the rates that certain genes are expressed, and to link these patterns with disease or other health outcomes.
“Biological data can be very noisy, in part due to how we collect and store this information, and the analytical methods we use make statistical assumptions about the data that, if not met, can impact interpretation,” said Konnaris, a doctoral student in bioinformatics and genomics. “Sometimes the tools we have to analyze these imperfect, messy data are heavily impacted by the noise, or the data don’t actually meet the assumptions, which can mean the tools might not answer the intended research question or give false results, fail to generalize to new data, or fail to reproduce the results that the creators intended.”
Konnaris works with his faculty advisers, Justin Silverman, assistant professor of information sciences and technology in the College of Information Sciences and Technology, and Nicole Lazar, professor and head of the Department of Statistics in the Eberly College of Science, to analyze and refine new computational genomic tools, many of which are grounded in statistical learning, machine learning or other artificial intelligence, to improve researchers’ ability to answer biological questions.
“We consider a research question, explore how current tools may or may not work as intended, and then refine existing tools or even develop something new that addresses the question,” Konnaris said. "We explore the theoretical and mathematical foundations of analytic methods, take into account the limitations of the observed data, and consider how the assumptions made about the system impact our ability to answer a question. Ultimately, we hope to develop a standard of practice that is rigorous and reproducible, enabling other researchers to develop more appropriate and effective tools.”
Konnaris originally set out to pursue a career in sports medicine and exercise physiology, but during clinical care and molecular wet-lab experiences, he noticed that there were opportunities to make the medical sciences more consistent and rigorous. His focus on analytic and computational technology development led him to Penn State.
“New computational tools have allowed us to propose analytical methods for analysis that were previously unimaginable, which is really exciting,” he said. “I saw the opportunity to make a huge change in the development of biological tools. Currently, there is a lot of interest in leveraging computational technology and AI tools to scale up analyses and make bigger and bigger models, and I am interested in making sure we do this in a rigorous way, grounding our analyses in statistics. Sometimes simpler tools can be more effective models. We hope our work helps researchers pick the right analytic tools for their goals.”
Editor's Note: This story is part of a larger feature about artificial intelligence developed for the Winter 2026 issue of the Eberly College of Science Science Journal.