CRISPR-Cas systems have revolutionized biology by allowing researchers to investigate how genes function. However, running CRISPR screens can exceed $10,000 per experiment and take months.
Following a weekend of intense computation, the winners of this year’s Ashby Prize Hackathon made a step toward a smoother, less expensive process. The Ashby Prize Hackathon is an annual event sponsored by the Center for Artificial Intelligence Innovation (CAII) in the National Center for Supercomputing Applications (NCSA) at the University of Illinois, Urbana-Champaign.
Over the weekend of October 24 and 25, “hackers” explored whether large language models could help predict the outcomes of these screens computationally, potentially helping scientists prioritize which experiments to perform, speeding up biological discovery.
Competitors built on prior work and a benchmark data set and, in just 15 hours, produced viable results.
Team 1, University of Illinois Ph.D. students Jiahua Dong and Anqi Tan, solution rose to the top of those results. With Dong as a computer scientist and Tan, a bioengineer, they checked all the boxes.
“They were able to identify all of the right challenges and they were able to address these challenges in their solution in unique and interesting ways,” said Assistant Director of NCSA Dr. Volodymyr Kindratenko.
With only two members, Team 1 was an oddity. Almost every other group was composed of three or more, but Dong and Tan considered it an advantage.
“Last year, we joined a five-member group. We spent almost the whole weekend discussing—a lot of talking and disagreeing,” Tan said. “I think two people are more manageable. We know our strengths, and we can focus on what we are good at.”
A big factor in the team’s success was discovering that the dataset provided wasn’t balanced — all the test examples had the same label (“true”). That means the data didn’t include both positive and negative cases (for example, “this gene affects the outcome” and “this gene doesn’t”). Because of that, any model would appear to be 100% accurate.
“The strategy we had was that we tried to decouple the problem into different parts, so classification and reasoning we treated as two different problems,” Dong said.
This year’s hackathon topic was uniquely difficult, according to organizers of the event.
At the beginning of the weekend, there were 19 teams; by the end, only 11, with the challenging hackathon assignment very possibly being to blame.
“That speaks for the complexity of this challenge,” Kindratenko said.
Per its name, participants were incentivized by the chance to win the Ashby Prize, established by Illinois Computer Science alumnus Steven Ashby and his wife, Beth. Ashby, Director of the U.S. Department of Energy’s Pacific Northwest National Laboratory, has long championed the transformative potential of high-performance computing in advancing research frontiers. With the esteemed first-place honor came $3,000, with second and third-place teams receiving $1,500 and $750, respectively.
Although for some participants, it only took a new opportunity to test their skills for them to show up.
“I don’t have much experience with CRISPR specifically, but I thought it would be a really good opportunity,” said Kavya Puranam, a first-year PhD student in informatics and member of team 14, who received an honorable mention. “Even if I didn’t get any results out of it, it was still a win for me just to get that kind of experience.”
Emine Senkardesler, a second-year PhD student in informatics and another member of Team 14, has been using AI essentially since it hit the market, but says that this was a new approach she hadn’t experienced.
“Of course, we use AI models, but not like this. Not like a prompting engineering type of thing,” Senkardesler said. “Like seeing how we can send the request to the models and seeing how it can do. It is actually very impressive.”
Under the guidance of Dr. Volodymyr Kindratenko, Dr. Shirui Liu and PhD student Dou Kwark from NCSA, and with technical support from Rohan Marwaha and Abhijith Nagarajan, these teams were powered by NCSA’s Computational Systems.
Teams presented their final solutions last Wednesday to a panel of judges composed of experts in computational biology, artificial intelligence, and high-performance computing. Entries were evaluated on innovation, computational efficiency, scientific relevance, and clarity of presentation.
NCSA Director Dr. William Gropp presented certificates to the winning teams at a recognition event on November 5, 2025, commending participants for their innovative thinking and dedication to advancing scientific computing.
“Even for the teams that didn’t finish, I think taking on the challenge itself is one of the most rewarding things,” Gropp said. “Everyone who took part in this, really, has excelled. We’re thrilled to be able to make that opportunity available.”
To see the winners click HERE