Samry Hehn [SH]: First, congratulations on your recent publication, “A Deep Learning Approach to Antibiotic Discovery.” Could you start by telling us a little bit about what this work is about, and the role you played in it?
Wengong Jin [WJ]: Thank You! This project is about using generative AI to design Novel Antibiotics. And this is important because of the global epidemic of micro bacterial resistance. A lot of antibiotics in clinics are no longer working due to bacteria evolving leading to developed resistance. We are short of antibiotics right now.
A lot of people when they contract ‘super bugs, like pneumonia, many die because of these bacterial infections. Antibiotics discovery is hard—in the past there was a long period of antibiotic discovery void. For example, a few decades ago, there was no new discovery of antibiotics, just improvements made to the antibiotic classes that already existed. Because antibacterial discovery is so challenging we choose a different route, instead of doing experimental screening we turned to AI method.
Generative AI models can automatically generate new compounds with exact properties. So, in this work I developed a generative AI framework based on my previous work that can automatically help create new compounds. We used that to propose compounds that are predicated to already possess compounds that have high antibacterial activity against two bacterial screens—those being Staph (MRSA one of the resistant strains of Staph) and Gonorrhea. Both strains are listed by WHO as urgent threat. So, we really need new antibiotics for these strains.
My role involves designing algorithms and setting up the computational screening pipeline as well as working closely with the experimental collaborators to prioritize what compounds to synthesize and test. My collaborators validated the efficacy of the design compound by AI models, and we found that it works very well both invitro and in vivo.
AI will automate a lot of the drug discovery, at least the computational aspect. Right now, we rely on humans to test the samples, maybe in the future we could have automated cloud labs where AI can test the molecules that they design.
Assistant Professor Wengong Jin
SH: Every big project comes with its hurdles. What did you find to be the most challenging part of this work?
WJ: Yes, well of course experimental testing is hard. However, my role is the computational aspect. As I touched on previously, the evolution of bacteria makes it challenging as they develop resistance, and we have a list of evolved strains we want to target. We found it relatively challenging to kill all the strains, but through generative AI and expanded optimization we managed to find compounds that can cure most of them.
Another issue we ran into was the synthesization of the compounds. In the first round, we didn’t take synthesizability into account much—we used a synthesizability score, and it wasn’t that great. So, we had 100 compounds and asked a company, “Hey, can you synthesize these for us?” Later, we found out that out of the 100, only two could be synthesized. Thankfully, one of them worked! If it hadn’t, the project would have died in the beginning.
SH: Wow! Talk about cutting it close! With that being the hurdle, you and the team overcame what would you say was the most rewarding part of this research?
WJ: I feel most proud of being one of the first groups that successfully used generative AI models to design new antibiotics. It was unprecedented. The majority of people in the past used virtual screening, fishing compounds from existing libraries to see if there is anything that will work. Unfortunately, many of them are suboptimal, so you need generative AI models to push the frontier one step further and to look for novel compounds that are very important.
SH: And looking ahead, AI is becoming a huge force in science and medicine. How do you see AI shaping the future of drug discovery?
WJ: The future will be revolutionized by AI by massive adoption of Generative AI models. AI will automate a lot of the drug discovery, at least the computational aspect. Right now, we rely on humans to test the samples, maybe in the future we could have automated cloud labs where AI can test the molecules that they design and then it will be a fully closed loop and make it efficient.
SH: Your work really sits at the intersection of AI and drug discovery. That’s a space where the next generation of scientists will need to be strong. From your perspective, how should we train young researchers so they can thrive and be productive in this kind of cross-disciplinary environment?
WJ: Right, I believe it begins with how to build a research group. For example, I come from a computational background with a PhD in computer science. If I build my group with just folks from computational backgrounds, it’s not going to work. We can’t test the compounds. We won’t be able to validate whether the compounds designed by AI really work.
So, I think it’s important for all researchers in AI to mingle and collaborate with researchers in the field. They need to work alongside one another to create a system that can take projects from AI to molecules in a closed loop, from designing to testing. This will allow young researchers to broaden their publishing capabilities. Rather than being tied to publishing only with computational publishers, they can expand to publishers like Nature and Science.
And as leaders, it’s our responsibility to build those bridges so computational folks can work closely with experimentalist folks.
SH: Building on that, what directions in your own research are you most excited about for the future?
WJ: I’m excited to develop an adjunct AI framework for drug discovery. We are in the process of building an adjunct AI framework not only for drug discovery but also for material science, DNA and RNA design, and many other areas. There are many f ields that can be transformed with an adjunctive AI framework. I feel this is going to be the future, and I am very excited about that.
SH: Here at the Barnett Institute, we’re always looking at how we can support innovative science. How has the Institute helped your research and your lab so far, and where do you see opportunities for it to play an even greater role in the future?
WJ: Barnett has been an incredible environment for supporting this interdisciplinary science. Having an institute with a combination of experimental researchers and AI researchers, like me, allows us all—chemists, biologists, computation folks—to come together to share ideas and work across traditional boundaries.
In the future, I think it would be innovative to have a sort of spinoff lab from Barnett to accelerate innovation, so we could bring our trials to the clinic. Right now, we are only testing in vitro.
SH: Many of our students and early-career scientists will be following this interview. If you could share one piece of advice with them about pursuing research, what would it be?
WJ: If you are AI folk, think about how you can make an impact… pick the right problems and find good collaborators that can bring your algorithm to impact. And I encourage experimentalists to work closely with AI folk, as it will accelerate your scientific research.
