Chris Seidel: The data sets are large and unwieldy. They often represent many tens of thousands of measurements, or dimensions of a system. AI and data science provide methods for extracting patterns, information, and features from highly dimensional data. The other challenge is more traditional and involves bringing hypotheses to evaluate against the data.

In terms of AI development, what are some of the most promising applications you see emerging within the field of computational biology in the next 5 to 10 years?

Sean McKinney: I do not think it is possible to see five years into the future right now, as things are moving too rapidly. Efforts so far have been largely limited to looking at 2D images, but newer developments make much better use of a third dimension, whether it be time or depth. This will enable researchers to look more closely at behavioral or other dynamic processes.

Jay Unruh: In the short span of a decade or two, biology will shift away from humans finding insights using direct observations to humans training computers to make those observations and us learning to draw insights from what the computer learned.

Chris Seidel: Just like ChatGPT understands language, we will have models that can understand DNA and protein sequences. This will help us understand biological molecules, and generate new ones with novel properties, enabling big advancements in synthetic biology and medicine. We'll also get better at understanding cells as systems.

What collaborations or interdisciplinary approaches are most important in advancing the research you’re involved in, and how does your team interact with other departments at Stowers Institute?

Jay Unruh: One of the biggest strengths of the Stowers Technology Centers is collaboration. This is especially true for our department. Because we work on so many kinds of data, we rely heavily on others, from curating data sets to evaluating the accuracy of results. AI can predict all sorts of things but until you have a plan to test those predictions they can’t make an impact on our scientific understanding.

Sean McKinney: We collaborate extensively with the Light and Electron Microscopy teams who help users design and run experiments using microscopy that can actually be quantified by our team.

Chris Seidel: We work one-on-one with researchers in the PI labs to bridge the gap between experimentation and computation, always with an eye towards what the next experiment might be. We especially like to work on open-ended exploratory projects to help solve scientific problems.

Can you share an example of a recent project or discovery that has benefited from the combination of big data, AI, and genomics? How has this impacted scientific understanding or potential applications in medicine?

Sean McKinney: We worked with the Gerton Lab to identify individual chromosomes in microscopy data and quantify the amount of a particular type of DNA called ribosomal DNA (rDNA) in hundreds of them. By analyzing chromosome sets from family members, we were able to show that these rDNA amounts are inherited.

Jay Unruh: In a recent project, I worked with Bret Redwine from Custom Protein Resources to design a small protein, or peptide, that would block an immune signal for the Li Lab. Our hope is that the peptide inhibitor could one day be used clinically for cancer treatment.

Chris Seidel: Advances in DNA sequencing recently enabled the complete sequence of the human genome, giving us insight into long overlooked areas. This has helped us understand how chromosomal structure can change and has allowed us to quantify differences in these regions across populations.

Are there any new instruments and/or technology that you are excited about?

Sean McKinney: We are working in a bizarre time when our field, though academic, is being advanced more by AI powerhouses like Google and Facebook than image processing academicians. It is exciting for me to just check what new tools are available every day.

Jay Unruh: As Sean mentioned, new tools are released every day. With modern AI tools, it is becoming possible to sequence the genome of an organism with a unique trait and propose the genes and molecules responsible for that trait.

Chris Seidel: The Systems Mass Spectrometry team has new machines that can rapidly quantify proteins in many samples. I'm curious about the large numbers of data sets that can be generated in this way and what we can learn from them. They also have machines for quantifying products of metabolism. I'm excited about the idea of combining genomic data with metabolomic data.