AI Agent & Copilot Podcast: Nandita Puri on How AI Is Revolutionizing Drug Discovery at Georgia Tech

**Giuseppe Ianni** (0:06)
Welcome to another episode of the AI Agent & Copilot Podcast, where we sit down and talk with influential leaders across the industry and talk about the opportunities, the impacts and the outcomes within AI, right? Today, I'm joined by Nandita Puri. She's a PhD researcher at Georgia Tech and analyst as well. So I wanted to welcome you, Nandita.

**Nandita Puri** (0:26)
Hi, thank you for having me over your podcast.

**Giuseppe Ianni** (0:29)
Yeah, I think it's exciting times, right? And I think as you are working at the forefront of drug discovery, a lot is changing, right? AI is accelerating the work across every industry, but specifically around drug discovery, right? And it's not just necessarily about efficiency, but it's also about possibility, right? And as we kind of make that jump from lab first to AI first, I know we had caught up a little bit for decades, drug discovery has been heavily dominated by what lab experimentation, and maybe you can talk with the audience about that. And now we see AI changing the starting point of drug discovery. What is your take on that?

**Nandita Puri** (1:09)
So I will just give a little bit of introduction about myself. I know what I do. I'm a third year PhD student at Georgia Tech, at Dr. Mekshen Lab.
I am doing my PhD in the intersection of bioinformatics and biochemistry. My job in my lab is to develop, is to study the interactions of proteins and lipids and to generate new lipids for autoimmune diseases, lipids as new therapeutic class for autoimmune diseases or other diseases where it's used in lipid metabolism signaling pathways. Yeah. So traditionally, a lot of these things have, I mean, people make hypotheses on pen and paper, or, you know, it's a more creative, it's a more discussion way of doing research work, where they would go read research papers, jot it down, write down the analysis, come out with the answers, and then take it to wet lab to validate it. There are also a lot of public databases, but it's very fragmented public databases, and the data is huge. It's huge. Which, I mean, a normal human being cannot analyze the whole huge database. So in recent time, due to the computing power that we have, I mean, increased computing power and the new algorithms that are coming into the market for AI, we have been able to analyze a lot more data using pattern matching algorithms or machine learning algorithms or neural networking algorithms. And using that pattern matching data, we are able to predict the future, or we are able to give an answer to a question or we are able to generate new therapeutics using that.
And so, yeah, so previously it just used to be wet lab, and now I see a healthy combination of 50% dry lab, which is the AI hypothesis algorithm building, and applying that onto the wet lab where the wet lab scientists go to the lab and validate our models.

**Giuseppe Ianni** (3:34)
Gotcha.

**Nandita Puri** (3:34)
So, yeah, that's a new trend coming up now in universities.

**Giuseppe Ianni** (3:38)
What do you see as the biggest advantage to that when we look at AI coming in more early stage in this process?

**Nandita Puri** (3:48)
I would say that, for example, each experiment takes about $10,000, $12,000, depending upon what the experiment is, you know, and that those experiments are done without full knowledge of the background knowledge, the background knowledge or background signals, background signals that is being given by databases.
So, the probability of an experiment failing is pretty high because we do not know the whole background knowledge, we do not know the signals that the data is giving, and we are taking it straight away to the wet lab. But now what AI engineers are coming in is that they are consolidating the whole database, picking up the signals from those database, telling us, telling the scientists like what genes, what drug modalities that you need to use, or simple just disconnecting the whole database. That is leveraging us to develop hypothesis faster. And the more, the faster you develop a hypothesis, and the more number of hypotheses that you can generate, the faster you can validate that in the lab, where you can also cut down the cost of experiments and also the probability of failing an experiment goes down.

**Giuseppe Ianni** (5:26)
Yep.
So a higher success rate definitely comes with it. Is it also helping them? As you explained that too, for me, I feel like it also gives the researchers better data to ask better questions.

**Nandita Puri** (5:38)
Yes.
Yes. Now that we have better data, we can actually ask better questions, and we would get better answers too for those questions. Yes.