A Seattle startup company has inked a deal with Eli Lilly to develop AI powered cancer treatments. The team at Lila Biologics says they’re pioneering the translation of AI design proteins for therapeutic applications. Anindya Roy is the company’s co-founder and chief scientist. He told KUOW’s Paige Browning about their work.
This interview has been edited for clarity.
Paige Browning: Tell us about Lila Biologics. You spun out of UW Professor David Baker’s protein design lab. What’s Lila’s origin story?
Anindya Roy: I moved to David Baker’s group as a postdoctoral scientist, where I was working on some of the molecules that we are currently developing at Lila. It is an absolutely fantastic place to work. It was one of the coolest experiences of my career.
The Institute for Protein Design has a program called the Translational Investigator Program, which incubates promising technologies before it spins them out. I was part of that program for four or five years where I was generating some of the translational data. I met Jake Kraft, the CEO of Lila Biologics, at IPD, and we decided to team up in 2023 to spin out Lila.
You got a huge boost recently, a collaboration with Eli Lilly, one of the world’s largest pharmaceutical companies. What are you hoping to achieve together, and what’s your timeline?
The current collaboration is one year, and then there are other targets that we can work on. We are really excited to be partnering with Lilly, mainly because, as you mentioned, it is one of the top pharma companies in the US. We are excited to learn from each other, as well as leverage their amazing clinical developmental team to actually develop medicine for patients who don’t have that many options currently.
You are using artificial intelligence and machine learning to create cancer treatments. What exactly are you developing?
Lila Biologics is a pre-clinical stage company. We use machine learning to design novel drugs. We have mainly two different interests. One is to develop targeted radiotherapy to treat solid tumors, and the second is developing long acting injectables for lung and heart diseases. What I mean by long acting injectables is something that you take every three or six months.
Tell me a little bit more about the type of tumors that you are focusing on.
We have a wide variety of solid tumors that we are going for, lung cancer, ovarian cancer, and pancreatic cancer. That’s something that we are really focused on.
And tell me a little bit about the partnership you have with Eli Lilly. What are you creating there when it comes to cancers?
The collaboration is mainly centered around targeted radiotherapy for treating solid tumors, and it’s a multi-target research collaboration. Lila Biologics is responsible for giving Lilly a development candidate, which is basically an optimized drug molecule that is ready for FDA filing. Lilly will take over after we give them the optimized molecule for the clinical development and taking those molecules through clinical trials.
Why use AI for this? What edge is that giving you, or what opportunities does it have that human intelligence can’t accomplish?
In the last couple of years, artificial intelligence has fundamentally changed how we actually design proteins. For example, in last five years, the success rate of designing protein in the computer has gone from around one to 2% to 10% or more. With that unprecedented success rate, we do believe we can bring a lot of drugs needed for the patients, especially for cancer and cardiovascular diseases.
In general, drug design is a very, very difficult problem, and it has really, really high failure rates. So, for example, 90% of the drugs that actually enter the clinic actually fail, mainly due to you cannot make them in scale, or some toxicity issues. When we first started Lila, we thought we can take a holistic approach, where we can actually include some of this downstream risk in the computational design part. So, we asked, can machine learning help us designing proteins that scale well? Meaning, can we make them in large scale, or they’re stable on the benchtop for months, so we don’t face those costly downstream failures? And so far, it’s looking really promising.
When did you realize you might be able to use machine learning and AI to treat cancer?
When we actually looked at this problem, we were thinking whether we can actually increase the clinical success rate. That has been one of the main bottlenecks of drug design. As I mentioned before, 90% of the drugs that actually enter the clinic fail. So, we are really hoping we can actually change that in next five to 10 years, where you can actually confidently predict the clinical properties of a molecule. In other words, what I’m trying to say is that can you predict how a molecule will behave in a living system. And if we can do that confidently, that will increase the success rate of drug development. And we are really optimistic, and we’ll see how it turns out in the next five to 10 years.
Beyond treating hard to tackle tumors at Lila, are there other challenges you hope to take on in the future?
Yeah. It is a really difficult problem to predict how a molecule will behave in a living system. Meaning, we are really good at designing molecules that behave in a certain way, bind to a protein in a certain way, but the moment you try to put that molecule in a human, it’s really hard to predict how that molecule will behave, or whether the molecule is going to the place of the disease, or the tissue of the disease. And that is one of the main reasons there is a 90% failure in drug development.
I think the whole field is moving towards this predictability of biological properties of a molecule, where you can actually predict how this molecule will behave in a human system, or how long it will stay in the body. I think when the computational tools become good enough, when we can predict these properties really well, I think that’s where the fun begins, and we can actually generate molecules with a really high success rate in a really short period of time.
Listen to the interview by clicking the play button above.
