Leveraging AI/ML to Reduce Risk in Drug Development: Part One

At AAPS PharmSci 360 2025, held from Nov. 9–12 in San Antonio, Texas, Pharmaceutical Technology® Group spoke with Sanjay Konagurthu, PhD, senior director, Science and Innovation, Thermo Fisher Scientific, about the conference presentation “AI/ML Model Informed Early-Stage Drug Development.”

In the above interview—the first in a two-part installment—Dr. Konagurthu discusses the potential of predictive modeling using artificial intelligence (AI) and machine learning (ML).

“The big advantage is time and cost savings and resource savings for the industry and our customers,” Konagurthu says. “Time is money, literally speaking, and what we bring to the table is reducing risk in product development, saving time and resources, minimizing API consumption. It's an ultra-material-sparing approach by leveraging some of these in-silico modeling tools like AI, leveraging AI/ML for supporting product development and formulation development.”

How can AI/ML improve early-stage drug development?

In particular, Konagurthu says, AI/ML can help sift through a field of drug candidates in which the percentage that are poorly soluble has increased in the past several decades from 30% to as high as 90%.

“It's a challenging environment and the probability of success of getting a poorly soluble NCE [new chemical entity] to a commercial product is pretty low,” he says. “It's way less than 10%, probably more like 5%. So, by leveraging some of these AI/ML tools that we've developed, we aim and strive to reduce the risk in product development.”

Check back for Part Two of PharmTech Group’s interview with Konagurthu, in which he drills down further on molecular dynamics simulations and quantum mechanical calculations as tools to be deployed beyond simple screening.

Click here for all of our AAPS PharmSci 360 coverage.

Transcript

Editor's note: This transcript is a lightly edited rendering of the original audio/video content. It may contain errors, informal language, or omissions as spoken in the original recording.

Hello, I'm Sanjay Konagurthu. I'm Senior Director of Science and Innovation at Thermo Fisher. I've been in the industry going on now 27 years.

Background, I'm a chemical engineer by education, PhD. My experience, broadly speaking, is in small-molecule oral drug delivery areas like solubility, bioenhancements, spray drying, hot-melt extrusion, nanotechnology, as well as modified control release, complex dosage forms, etc., ranging from early development through commercialization.

I also lead a predictive modeling and simulation team at Thermo Fisher. We are a global team. We deal with the predictive modeling and simulations for drug product development.

That's a great question. I like to tell people when I first started working in this industry over 20-plus years ago, the number of candidates that were poorly soluble were estimated to be 30% or so. Now, depending on the source that people quote, 70% to 90% of molecules are poorly soluble, so it's more than doubled, if not tripled.

It's a challenging environment and the probability of success of getting a poorly soluble NCE to a commercial product is pretty low. It's way less than 10%, probably more like 5%. So by leveraging some of these AI/ML tools that we've developed, we aim and strive to reduce the risk in product development.

We can get to accelerating the early phase especially by leveraging these in-silico AI/ML tools for selection of technologies. What's the best technology that might work for any given molecule? And it's all dependent on the type of molecule. Every molecule is unique.

And also in formulation selection, for example, we deal with complex modalities like protacs, which are on the larger end of the spectrum, and the traditional small molecules. And all these molecules violate, typically, Lipinski's rule of five. So they're literally the brick dust, greaseball-type molecules. By leveraging these AI/ML tools we again, as I said, reduce the risk in product development, get to the milestones faster, and shorten timelines, etc.

The big advantage is time and cost savings and resource savings, for the industry and our customers. Time is money, literally speaking, and what we bring to the table is reducing risk in product development, saving time and resources, minimizing API consumption. It's an ultra-material sparing approach by leveraging some of these in-silico modeling tools like AI, leveraging AI/ML for supporting product development and formulation development.

Every API, as I mentioned, is unique. So your structural features are very important, obviously the chemical structure, also fundamental properties like the melting point which have bearing on the crystal lattice, [the] energy of the molecule [is] important. Other properties like pKa, LogP, LogD, all of those are important.

Also, things like permeability are important. So typically we try to get data measured in in-vitro assays like KCO2 or MDCK, and we can leverage that to predict what's happening in the human body. These are some of the features that are important in some of these calculations that we do.