January 7, 2025
By Holly Smith
“The goal of the fellowship was to train students for artificial intelligence (AI) and machine learning (ML)—some things that are very new, even new to me. You definitely need these techniques to make sense of patterns that humans can't find on their own,” shared Professor Esmael Haddadian, creator of the AI and ML in Biology Fellowship.
This summer the Biological Sciences Division (BSD) and the Research Computing Center (RCC) cohosted the AI and ML in Biology Fellowship. To join the fellowship, students applied and answered questions about their research and areas of expertise. The applications were then sent to research groups interested in the fellowship who selected the fellows. “I applied because I’m a computer science and biology double major, so an opportunity to work on AI and ML in a biology lab sounded amazing. I was hoping to get some hands-on experience in applying machine learning concepts to biology datasets and to gain some insight into the field that combines these two fields,” detailed Alissa Domenig, one of the summer fellows.
The fellowship had several components. It started with two weeks of lectures and workshops conducted by RCC computational scientists, NVIDIA, and Argonne National Lab, who introduced the talented undergraduate students to the world of AI applications in the biological sciences.
“It’s always exciting to interact with such bright students because they raise questions that make you really think. My sessions were very interactive so it was a very fun learning experience for both parties,” shared Debasmita Samaddar, RCC Computational Scientist. “Working alongside peers who were equally passionate about exploring AI’s role in life sciences made each session inspiring and challenging,” agreed Mohsen Zand, RCC Computational Scientist.
“The workshops in AI and ML methods were extremely interesting and a big step for me in breaking down higher-level concepts. I learned a lot about the AI-driven systems that I took for granted before this experience, and the kind of implementation/career paths that are available for data science in the health science domain. Having multiple lecturers from different sources was all informative and directly helpful with my research project. This fellowship has also allowed me to progress in terms of my mentorship, which otherwise would likely not have been possible,” enthused summer fellow Michelangelo Pagan.
Next, the fellows worked in a research group for eight weeks led by a University of Chicago principal investigator (PI). To conclude the fellowship, the fellows presented their work to each other, and all PIs involved.
The fellows’ projects had a wide range of topics. For example, Jhan Liufu worked on manmade technologies that translate brain activities into human interpretable concepts like the English language and movements. “Brain-computer interface (BCI) technologies have existed for a while, but they fail to work across different people as the ways our exact brain cells fire can vary. To work on this, there is a process of individual calibration. Right now, this process takes a lot of time and a lot of human cost. My project was to work towards an ML learning technique to streamline calibration,” explained Liufu.
Domenig split her time between studying how environmental variables and community functions relate to distinct taxonomic groupings using a variety of dimensionality reduction techniques and exploring building neural networks to optimize community composition in soil for degrading BPA. “The idea is to find a way to predict which microbial community composition is best suited for degrading BPA at various concentrations,” described Domenig regarding her second project.
Pagan looked to predict changes in physiological frailty over time. “Generally speaking, the project looked to evaluate ML models that are most suitable for predicting changes in health for older adult populations given simple accelerometer data-collection, which significantly reduces the time burden geriatricians face in doing clinical evaluations of their patients and could streamline the identification of preventative measures without the need for additional in-patient screens,” divulged Pagan.
What everyone found rewarding regarding the fellowship also varied. “The interaction with and amongst the students was the most rewarding,” said Haddadian, “As well as learning about the interesting projects they did.”
“The most rewarding experience has been building up the skills required to work with ML on huge datasets, from running jobs on the midway clusters to fine-tuning hyperparameters and optimizing model performance. I had never worked with biology datasets in the context of ML before, so it has been exciting discovering all the possibilities that merge biology and ML and what they can offer in understanding biological phenomena and advancing research in the field,” explained Domenig.
Haddadian and the RCC are grateful to the University for funding this fellowship and look forward to repeating and expanding the program in future years.