We could have also titled this At the Lab episode “20,000 leagues under the genome.” This week, you’ll hear all about SQUID, the new artificial intelligence tool from CSHL Assistant Professor Peter Koo and Associate Professor Justin Kinney. How can this AI help scientists crack the mysteries of the genome? Tune in to find out.
Read the related story: SQUID pries open AI black box
Transcript
Nick Fiore: You’re now At the Lab with Cold Spring Harbor Laboratory. My name is Nick Fiore, and this week At the Lab, “AI SQUID.”
{Water bubbles.}
NF: Imagine a black box resting on the seafloor. What’s inside? Your guess is as good as mine.
NF: In a way, artificial intelligence is similar. As advanced as AI has become, today’s computer scientists have very little understanding of its inner workings. That goes for today’s most popular AIs—the image recognition platforms and large language models—as well as more specialized applications.
NF: Computational biologists are now using AI models to try and better understand health and disease. These models can analyze a genome and spit out predictions about the function of different parts of that genome or individual genetic mutations.
NF: At least, that’s the idea. But there’s a problem. CSHL Assistant Professor Peter Koo explains:
Peter Koo: The tools that people use to try to understand these models have been largely coming from other fields like computer vision or natural language processing. While they can be useful, they’re not optimal [for genomics].
NF: Hence, the black box has remained, for the most part, tightly sealed. But wait … what’s that in the distance?
{Water bubbles increasingly louder with splashes.}
NF: Here comes the latest AI model from Koo and CSHL Associate Professor Justin Kinney. Its name? SQUID!
NF: SQUID stands for Surrogate Quantitative Interpretability for Deepnets. Don’t worry about what that means. What’s important is SQUID’s intended purpose—to pry open the black box of genomic AI models.
{Metal wrenches.}
NF: In other words, it’s built to help biologists understand just how AI goes about analyzing the genome. From there, they can fish out an AI’s most accurate predictions from inside the computer world. If right about now, you’re picturing a squad of SQUID-like robots taking over biology, Kinney assures us that’s not the goal here.
Justin Kinney: In silico [virtual] experiments are no replacement for actual laboratory experiments. Nevertheless, they can be very informative. They can help scientists form hypotheses for how a particular region of the genome works or how a mutation might have a clinically relevant effect.
NF: And that could bring scientists closer to their true goals—understanding life at its most fundamental level, figuring out how it evolves and adapts, identifying the root causes of diseases and potential cures.
NF: Squids have a lot of arms—six to be exact, plus two tentacles. Likewise, the SQUID AI could have a number of promising applications … robotic cephalopods notwithstanding.
NF: Thanks for joining us this week At the Lab. If you like what you heard, please consider subscribing wherever you get your podcasts. You can also find more stories about AI and other fascinating topics at CSHL.edu. For Cold Spring Harbor Laboratory, I’m Nick Fiore, and I’ll see you next time At the Lab.