黑料吃瓜资源

Artificial intelligence has taken the world by storm. In biology, AI tools called deep neural networks (DNNs) have proven invaluable for predicting the results of genomic experiments. Their usefulness has these tools poised to set the stage for efficient, AI-guided research and potentially lifesaving discoveries鈥攊f scientists can work out the kinks.

鈥淩ight now, there are a lot of different AI tools where you鈥檒l give an input, and they鈥檒l give an output, but we don’t have a good way of assessing the certainty, or how confident they are, in their answers.鈥� explains 黑料吃瓜资源 (CSHL) Associate Professor Peter Koo. 鈥淭hey all come out in the same format, whether you鈥檙e using a large language model or DNNs used in genomics and other fields of biology.”

It鈥檚 one of the greatest challenges today鈥檚 researchers face. Now, Koo, former CSHL postdoc , and graduate student have devised a potential solution鈥擠EGU (Distilling Ensembles for Genomic Uncertainty-aware models). DNNs trained using DEGU are more efficient and more accurate in their predictions than those learning via standard methods.

鈥淲hen we want to make claims in biology, we don鈥檛 want to rely on a single model,鈥� Koo explains. 鈥淔or example, we might train 10 models, and we鈥檒l get predictions from each. Typically, we鈥檇 then use a method called deep ensemble learning to see where they agree and disagree. But handling 10 models and ensembles is challenging, especially as the model sizes grow. That鈥檚 where DEGU comes in.鈥�

is built on a previously developed method called 鈥渄eep ensemble distribution distillation,鈥� which focuses on learning a DNN鈥檚 overall distribution of predictions rather than its individual estimates. Regardless of how many models are used, DEGU distills the resulting ensembles down to one much more manageable tool. Koo, Zhou, and Rizzo found that models trained using this distillation process provided better predictions鈥攁nd better explanations for those predictions鈥� than those without it. They also required less power.

鈥淚nstead of needing to analyze 10 models at once, you鈥檙e working with a single model one-tenth the size with the same predictive capabilities,鈥� Rizzo explains. 鈥淎nd because you鈥檙e only working with one model, it鈥檚 easier to understand what鈥檚 driving its predictions and uncertainty.鈥�

The Koo lab is now working to improve DEGU鈥檚 efficiency and make it more accessible to researchers worldwide.

鈥淟ab experiments are expensive,鈥� Rizzo says. 鈥淚f we can make our models as reliable as possible for downstream applications, scientists will spend less time chasing predictions that a model isn鈥檛 even that confident about.鈥� Fewer wild goose chases and better bases for strong hypotheses鈥攖hat鈥檚 the goal. Now, we鈥檒l see how DEGU delivers.

Written by: Nick Wurm, Communications Specialist | [email protected] | 516-367-5940

Funding

National Human Genome Research Institute, National Institute of General Medical Sciences, National Institutes of Health

Citation

Zhou, J.,听et al., 鈥�Uncertainty-aware genomic deep learning with knowledge distillation鈥�,听npj听Artificial Intelligence,听January 7, 2026. DOI:听