At the Lab Season 1 Research Rewind: AI+

Transcript

Nick Fiore: You’re now At the Lab with Cold Spring Harbor Laboratory. I’m Nick Fiore.

Sara Giarnieri: My name is Sara Giarnieri.

Nick Wurm: I’m Nick Wurm.

Sam Diamond: My name is Sam Diamond.

NF: And this week At the Lab we’re rewinding several episodes from Season 1 that focus on AI.

{Music}

NF: Of course, that’s a big topic. But while the media focuses on chatbots and other internet applications, CSHL’s quantitative biologists are applying AI to problems in neuroscience, genomics, health care, and beyond. So strap in as our final Research Rewind is a deep dive.

{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.

{Music}

Sara Giarnieri: There’s a symbiotic relationship between neuroscientists and computer scientists worldwide. These specialists come together in a field called neuroAI. Their collaboration both aids our understanding of brain function and improves artificial intelligence.

SG: CSHL’s NeuroAI Scholars Program recruits young AI experts to work with neuroscientists on campus. Last summer, Long Island locals heard the inside scoop from one of the program’s recent participants.

SG: NeuroAI Scholar Kyle Daruwalla discussed his work during Cocktails & Chromosomes, our monthly science talk held at Industry Bar in Huntington, New York.

{A drink is poured.}

Kyle Daruwalla: All the great neuroscientists at Cold Spring Harbor—they provide the insights from biology that I use for my research, and I collaborate with them. And what I bring to the table is my background in computer science.

SG: Daruwalla trains AI based on how the human brain learns and adapts. Like the human brain, AI learns by making connections. But it doesn’t have the same neural patterns that make this process so efficient for us.

KD: Our brains are made up of cells, full neurons, and these neurons are connected to each other. And they’re not just connected to each other randomly. They have a very specific pattern to those connections. They’re very structured. And this is why our brains are able to do all the things we’re able to.

SG: The Human Genome Project took 13 years to complete because human biology is so complex. The same could be said for human intelligence—the product of millions of years of evolution. Now, scientists are looking for a way to help artificial intelligence catch up.

KD: We’re trying to port this idea of a genome—an idea of having a structured pattern as soon as we start out—to AI models.

SG: Herein lies the promise of neuroAI. With the help of neuroscience, AI could become easier to train and more energy efficient. That would make it more accessible for everyone. In turn, AI could teach us more about the human brain. This could lead to a better understanding of neurological conditions like Alzheimer’s, depression, autism, and more.

{Music}

Nick Wurm: What’s that smell?

{A nose sniffs the air.}

NW: It sounds funny, but in many ways, that question is central to the human experience. Imagine you’re in the wild. Something smells awful. It repulses you. You don’t eat it, and you live to smell another day.

NW: Now, that’s a basic example, but our odor palates can actually become quite refined. CSHL Associate Professor Saket Navlakha offers wine tasting as an analogy.

Saket Navlakha: People can’t discriminate between two red wines today and then can discriminate between them after three months of taking wine tasting classes. What is actually happening in your brain potentially that allows you to do that? We’re providing one answer to that question.

NW: Navlakha is a quantitative biologist. He looks at big questions in life and nature like complex math problems. And he uses computer science to seek elegant solutions.

NW: In this case, Navlakha and his team developed a computer model based on how fruit flies’ brains respond to certain odors.

{Flies buzz.}

NW: The team found some neurons in the fruit fly brain respond differently to two dissimilar odors but the same to similar scents. The researchers called these neurons reliable cells. This small group of cells helps the flies quickly distinguish between odors that are very different from one another—like rotten and fresh fruits.

NW: On the other hand, another much larger group of neurons responds more unpredictably when the fly encounters similar smells. The researchers called these neurons unreliable cells. They think these neurons might help us learn to identify specific scents—like the notes in a glass of wine.

{A glass is poured.}

NW: But in case you’re wondering, no, this research isn’t just for wine aficionados. Remember, we’re dealing with computer models here. And as with all computers, data comes in, data goes out. Navlakha explains:

SN: Maybe you don’t want a machine-learning model to represent the same input the same way every time. In more continual learning systems, variability could actually be useful.

NW: Want to make AI more reliable? Then it needs to learn to be more discerning. And here’s a great place to start.

{Music followed by chirping sounds}

Sam Diamond: Birds come in all shapes and sizes. Just this morning I saw a blue jay in one of the trees outside my office. Right now I see a flock of seagulls floating in the Harbor. And if I’m really lucky I might even spot one of the bald eagles that calls our campus home.

SD: But what makes these birds so different from one another? To find out, we sat down with CSHL Professor Adam Siepel.

Adam Siepel: Birds often display very pronounced morphological differences from one subspecies to the next. They often exhibit sexual selection for particular prominent coat color or song changes.

SD: Siepel is an expert in population genetics. Recently, his lab took an interest in seedeaters—finch-like birds from South America. Tens of thousands of generations ago, these birds all had identical genetic codes. That meant similar feathers and birdsongs. But since then, they’ve broken off into many species with different coats and calls. Siepel says the different species’ genes reflect these changes.

AS: When we sequence their genomes, their genomes are quite similar to one another. But they have local regions that are highly differentiated. These have been referred to as islands of differentiation.

SD: What causes these islands to emerge? One way they can come about is through physical separation. Birds may develop a new species when one group becomes separated by a large barrier, like a mountain range or an ocean. However, Siepel and his lab found that it wasn’t geographic boundaries that caused seedeaters to drift apart—it was something else.

AS: A selective sweep, when a group of organisms that carry some mutation that gives them an advantage over other organisms, rapidly become much more frequent—maybe because of a change of environment, a new predator, a new food.

SD: The sweep occurs when individual animals with a particular genetic variant begin to reproduce at higher rates. Siepel explains.

AS: In this case, we think the birds of the opposite sex found some aspect of that variant attractive, whether it’s coloration or song. And that helped push it to high frequency.

SD: From there, the variant individuals stay together, eventually forming a new species. Biologists are still trying to figure out why birds find certain colors, songs, or other variants more attractive. In the meantime, Siepel’s work suggests that the old saying holds true in evolution. Birds of a feather really do flock together.

{Music followed by heartbeat sounds}

Brianne Seviroli: When it comes to the heart, there are many mysteries. One in particular dates back centuries. It was stumbled on by none other than Leonardo da Vinci.

BS: About 500 years ago, da Vinci noticed a cobblestone-like pattern of muscles lining the inside wall of the heart. And for 500 years nobody knew why they were arranged like this … until now.

Hannah Meyer: Let me put it this way. They reduce the odds for heart disease.

BS: That’s Hannah Meyer, an assistant professor at Cold Spring Harbor Laboratory. Meyer took a holistic view to researching heart disease. She compared the organ’s genetic makeup with its phenotype—how it looks and behaves.

HM: I was very interested in organ function as a whole. Can we use the genetics and these phenotypes that we can extract to understand more about the physiology of an organ, both in health and disease? So, we teamed up with bioengineers and with clinical researchers to try and understand a small phenotype in the heart from the perspective of how does it influence the function of the heart.

BS: Using special software, Meyer and her colleagues analyzed the clinical data of 25,000 patients in the UK Biobank. From here, the team was able to see how the heart wall’s cobblestone-like muscles, called trabeculae, work and develop.

BS: And guess what? It’s a lot like a golf ball turned inside-out. If you’re a golfer, you might know that the dimples on the ball reduce air resistance, helping it travel farther. Trabeculae cut down on fluid resistance. Hearts with the right kind of muscle pattern are able to pump blood with less resistance. Their trabeculae branch out in a shape resembling that of the heart. And Meyer’s team found that patients who have this going on are at lower risk of heart failure.

BS: Meyer has since turned her data-driven research to cancer.

HM: Every piece of software that I’ve developed is on an open platform. There’s more than 20,000 people who’ve downloaded it.

BS: And that means that thanks to her, we may soon have a better understanding of why many other organs look the way they do. Eat your heart out, da Vinci.

{Music}

{Bats shriek and flap their wings.}

NW: Imagine somebody walked up to you and said that one word: bats.

NW: Now, you don’t really have to imagine much. That’s the actual origin story of a recent Cold Spring Harbor discovery. Here’s CSHL Professor Dick McCombie.

Dick McCombie: A student from a course we’ve been teaching since 1995 and who works at the American Museum of Natural History was collaborating with someone in my lab. She was visiting, and I always kid around with her. And I was pretty psyched up about bats at the time. And I pointed at her and said bats! And she said, “Dick, what the hell’s wrong with you?” I told her how interesting bats were. She said, “Have you met Nancy Simmons? She’s a bat researcher at the American Museum of Natural History.” She’s the one that got the bat samples from Belize.

NW: Before we get to Belize, let’s backtrack a bit. Bats: what makes them so interesting? For you and me, it might be their nocturnal habits or their appearance in popular folklore.

{A vampire laughs.}

NW: But for McCombie, it’s not their legends or lifestyles so much as their lifespans.

DM: There’s a general trend that the bigger the average body size of an animal, the longer they live. Bats are real outliers in that regard. Some species live far, far longer than would be expected based on their body size. And they apparently have a very low rate of cancer.

NW: McCombie and his colleagues wanted to find out why. Enter: the American Museum of Natural History team in Belize. They provided McCombie with DNA samples from two species of bats, the Jamaican fruit bat and Mesoamerican mustached bat.

NW: Back at CSHL, McCombie and his colleagues mapped the first-ever genome sequences for these two types of bats. When they compared the genomes to those of 15 other mammals, including other bats and humans, they found that the genes responsible for bats’ immune responses are dialed way down.

NW: As a result, their immune systems might work more quickly and precisely, lessening the amount of friendly fire on bats’ organs and tissues. That could help explain their longer lifespans and apparent resistance to cancer.

NW: McCombie’s colleagues hope their work will help provide new insights into the links between immunity, aging, and cancer. And to think, it all started with one word: bats.

{Music}

NF: Thanks again for listening to Season 1 of our new podcast, At the Lab. If you like what you heard, be sure to subscribe wherever you get your podcasts, and stay tuned for more breakthrough bioscience coming in 2025. Remember, you can also find us online at CSHL.edu. For Cold Spring Harbor Laboratory, I’m Nick Fiore, and I’ll see you next time At the Lab.