AI is designing drugs in months—not years. But the real bottleneck? Clinical validation. In this special episode of Below the Fold—recorded live at the Dubai AI Festival—Dr. Alex Aliper, CEO & President of Insilico Medicine, breaks down the final frontier in AI-powered drug discovery. From lab automation to quantum breakthroughs, Alex shares how AI is transforming every layer of pharmaceutical R&D—yet regulatory systems remain slow to catch up.
🎙️ Topics include:
In collaboration with Date With Tech, Launch Foundry // AI Product Studio, Trescon, and the Dubai AI Festival at DIFC Innovation Hub.
[00:00:00]
Hi, I Am Will Snow and I'll be your host today with Date With Tech in association with Trescon, Launch Foundry, and below the fold. We're at the Dubai Air Festival and today I'd like to welcome Dr. Alex Aliper, co-founder and president of IN Silico Medicine. Alex, welcome. Thank you. Great to be here.
Cool. Good to have you. So, Dr. Alex,~ uh,~ a few questions. ~Um, ~you know, we're discussing the future of medicine, AI's new Frontier and drug discovery. What, what first got you thinking that AI could actually help discover new medicines? ~Uh, ~was there a breakthrough moment? I think from a get go and,~ uh,~ we started back in 2014.
Uh, back then the AI was very nascent and Generat AI was very nascent, right? It's kind of the,~ uh. ~The year that the first papers started coming out and ~uh, ~when we started in biology, we started in [00:01:00] oncology specifically trying to identify reasons, molecular pathways that drive cancer growth, that drive the onset of this disease.
~Um, ~we wanted to modulate those pathways we identified and we. That's how we entered the pharmaceutical kind of realm of,~ um, uh, ~business. And we started talking to pharma companies and soon enough we realized that within pharma companies, big pharma companies, it's, you know, it's a great organization, very structured, but some, you know, processes can be like optimized.
Some departments are pretty disjointed for sure, and on average it moves pretty slow, right? So it takes. Like by different estimates from 10 to 17 years to get a single molecule approved. Wow. That's a long time. It's a long time. And it takes billions of dollars to get a single drug on the market. ~Um, ~so it's terribly slow and terribly expensive.
And we thought [00:02:00] that, you know, why don't we use the latest advances in ai, uh, in, in both biology, in chemistry, and see if we can make a difference if we can cover this. Entire drug discovery cycle and then go into drug development,~ um,~ with the help of AI and make it more efficient and fast. Understood.
Yeah. Amazing. It's, for me, it's always very interesting when I, I think about ai, I see a lot of it in the commercial advertising space. ~Um, ~but you know, the, from a medicine perspective and even like in the environmental side, we haven't heard so much. So it really inspires me when I see work coming out of the medical space,~ uh,~ using this advanced technology.
Yeah, I think AI is touching every industry these days. And,~ uh,~ medicine, healthcare is no exception. Uh, it already, I mean, AI made strides in, uh, diagnostics or diagnostic tools approved by FDA,~ um,~ and,~ uh,~ it really helps physicians today make, um. A [00:03:00] really highly accurate,~ uh,~ diagnosis and,~ uh,~ prescribed treatments, but necessarily on the beginning Yes.
Of this journey. So yeah, AI has a long way to go in healthcare for sure. And like you said, every industry is affected by this. How are medical institutions dealing with the rapid change? What's happening inside these organizations? Well, it depends on the institution. Some are more, um. Conservative, some are more kind of daring and innovative.
~Um, ~if we're talking about hospitals, I think Yeah. ~Uh, ~hospitals that are on the cutting age, they have adopted some of the ai,~ uh,~ technologies specifically within the domains of,~ um,~ medicine related to diagnostics, imaging. ~Uh, ~so imaging is super, like probably the first area. That was touched by AI in a meaningful way, ~ways, images.~
~Right. So, and imaging data. ~And that's why we have like first applications being in medicine, being approved related to imaging. I think, yeah, in healthcare, like hospital [00:04:00] setting, the adoption is, well, it'll take time,~ uh,~ but it's probably the right also approach because. When I think about AI in healthcare, I also think of, you know, potential like distance to HA to potential harmful event.
Like what are the risks of adopting AI too quickly? For sure. Yeah. So if it's, if it's distance to a harmful event, theoretical event is long, then it's okay to implement like faster. If it's shorter, then you need to take more time. Okay, that makes sense. Depending on the diagnosis and what the patient's dealing with.
Yeah. Cool. Okay. I understand that. That's, that's a smart way to do it. So we're still including the human on the high risk kind of side of things, and we're starting to introduce the tech on the more low, low risk side side of the business. Yeah, yeah, yeah. So, which makes sense. ~Uh, it, it, it's, it's one thing when you kind of, uh,~
~identify something within a CT scan. ~Right. And another thing, when you use AI to operate in, uh, a machine that performs surgery, for sure. Okay. Right. So. Makes sense. Makes perfect sense. Yeah. Perfect. ~Um, ~let's move on to the next question. With regards to AI drugs really, which is [00:05:00] what you're doing, how do you,~ um,~ get these into human trials?
What's the process involved in that? Is it a bit more tricky? Yeah, typically,~ uh,~ the process consists of,~ uh,~ several stages on the grand scheme of things. To get a drug approved, you need to complete drug discovery and drug development. Drug discovery is everything. ~Uh, ~what happens before drugs enters human clinical trials?
So it's prehuman. Okay. Stage,~ uh,~ and drug development is human clinical trials, phase one, phase two, phase three, four, and launch, commercial launch of the product,~ um,~ of, of a drug. ~Um, ~so drug discovery typically takes five years, and it includes in traditional industry, it includes,~ um,~ target identification.
So you need to understand. What is the origin? What is the causal reason why the disease happened?~ Um, ~so that's typically a protein target. Is a protein, okay. ~Um, ~that we want to modulate in the right way to change the trajectory or cure the disease. Once you identify this [00:06:00] protein, you need to generate a molecule, an actual drug that would modulate that protein in the right direction, either inhibit or modulate, ~uh.~
Up in terms of,~ uh,~ upregulation and,~ um,~ can be more kind of subtle mechanisms of,~ um,~ modulating a target. And once you identify this molecule, you optimize it for safety. Uh, so it must be, uh, very safe. It must be efficacious in animal models and cell-based models, organoids and. After you nominate your developmental candidate, the final molecule that you are sure works in animals and you are confident it will work in humans.
You go into IND enabling studies that prepare you for human clinical trials. So this is where discovery stage ends, well, IND enabling sometimes refer to as preclinical,~ um,~ stage of,~ uh,~ drug discovery. So that's the process. Typically, it takes five years. With the health of ai, we showed that we can do it in 18 months or [00:07:00] even faster.
Uh, for our lung fibrosis program, it was 18 months. Okay. And we did it on a fraction of a budget of traditional industry instead of hundreds of millions. We did it on like two and a half million. Incredible,~ uh,~ dollar budget. Yeah. Great. So it's, it's really bringing efficiencies in, in both,~ uh,~ a monetary aspect,~ um,~ as well as a time aspect.
Um, to the, to the industry. Yeah. Which is fascinating. It's, it's a very deep subject, one close to my heart. Love it. It is a rabbit hole. Yeah. Yeah, for sure. Yeah. ~Um, ~and I, I just think with these things, you usually have such a big team of engineers and,~ um,~ scientists working behind them. How do you keep innovation focused and grounded in the real, in real world results?
I think this framework around like actual medicines that we're discovering really helpful. Because ultimately every single effort funnels through this, uh, workflow. No matter if you're working in biology, try to identify the target or chemistry, generate the [00:08:00] molecule, or you're developing software that helps others, uh, discover targets and molecules.
We have software business as well. Yeah, we sell software to big pharma and academia and biotech. No matter what you do in the company. Ultimately you are contributing to the drug discovery, to the discovery of a medicine that will make a difference that will in people's lives. Right. So Absolutely. In patient's lives, and this is super motivating.
Uh, and it's like driving force Yeah. Behind like every single um, uh, employee we have. And, uh, in terms of innovation, that's also what. Fuels innovation. We want to make it more efficient. And in order for us to break those paradigms, we need to innovate. We need to invent, uh, new workflows, faster cycles, um, and innovate not only in terms of algorithm, but [00:09:00] also operational excellence, right?
So if efficiency, how do we move those experimental processes faster? 'cause in drug discovery, you, you cannot evaluate the quality of your. LLM output. Unless you go in the lab and physically make the molecule and test it, you, you need to spend the time and resources to go in the lab, synthesize a molecule, test it.
It's not Chad GPT, where you see for sure response. Makes sense. Sense. You need all the information you need, you need to experiment. Yeah. And that's,~ um,~ both a blessing and a curse. A blessing. Yeah. Because,~ uh,~ it also, you know, ensures that whoever. Goals were first have a lead in that, uh, domain. And we were fortunate to be a pioneer in this field.
But of course, it, uh, it is in a way disadvantages because,~ uh, yeah, ~you'll have to spend this time anyway no matter how, like what, what, what fancy algorithm you have, you need to validate it in the lab. Yeah, [00:10:00] you're basically generating the data from nothing, right? Yeah. So there's nothing to train it. ~Um, ~and that's what's so experimental about what you guys do.
Um, so it is, it's very much cutting edge, if not bleeding edge. I think something that you mentioned is really having people, having passion for what they're doing and how it is changing the world. Yeah. If you, if you could recommend, um, a piece of software or piece of advice for people starting off their journey into this healthcare, tech, um, industry, what would you, what would you say?
I mean, you can start with some of our papers, for instance, because we cover the entire cycle of drug discovery and software wise, we have a software for the full stack of, uh, drug discovery tasks from biology to chemistry. Uh, we have many software applications including automation,~ uh, piece ~with our automated lab, uh, full automated lab operated, um, autonomously.
~Um, ~so we can start there. ~Uh, ~start with some of our papers. Of course. I would encourage everyone to embrace,~ um,~ ai, embrace other [00:11:00] lambs. Try try them,~ uh,~ test them, challenge them, you know, get curious about them because they will be more and more,~ uh,~ powerful, more precise. We already can reason at a pretty good level.
And,~ uh,~ yeah, this will be. You know, ~um, ~integral part of our lives. Well, in a way it already is. Yeah. I love that. I love the fact that you're also mentioning like challenging it. Yeah. ~Um, ~I think that's really where excellence comes from. Yeah. ~Uh, ~which is fantastic. And, and I, I guess going back to the challenges, what is like one component of this whole,~ uh,~ phase that you would love to see sped up?
I mean, we
we're doing pretty good job on speeding up. Those phases that can be,~ um,~ improved in terms of efficiency on a discovery stage, uh, discovery stage. I think,~ uh,~ clinical trials are there for a reason, so we need to make sure that the drugs are really safe and [00:12:00] efficacious. So you cannot really like cut corners and bend the rules,~ uh,~ in a clinical trial for sure.
Uh, stage. In drug discovery, we're pretty efficient. I would say like areas that we can,~ uh,~ improve on would include faster chemical synthesis of the generated molecules. Faster testing. But this is where robotics come in, uh, come in to help us, right? So we have robotics lab enable both automated synthesis and automated screening of the molecules.
So automation is coming to drug discovery as well. And it is touching every industry. So I think it'll solve many of those challenges from the timelines. Yeah, we've definitely seen AI starting to merge into robotics. I think it'll be, uh, a big theme this year is obviously genive agents, but as we see that shifting into robotics,~ um,~ that's really where I think I see a larger benefit,~ um,~ for all industries.
Yep. Amazing. Cool. I, I guess the last question from my side is [00:13:00] just, uh, what's one thing,~ uh,~ that's happening out there with AI and stuff that the public,~ um,~ the public perception or the public people don't actually know about, but they should probably be aware of? I think there are a few, like a few things, right?
So, ~um, ~I think public is not yet ready to the speed of innovation in the field of robotics. I think it's gonna be. Coming to a reality sooner than most of the public thinks.~ Uh, ~I also think that,~ uh,~ related to new technologies, right? So there is a big hope on quantum computing. Yeah. ~Uh, ~it, it has kind of it's ups and downs in terms of like publicity, right?
So, uh, but it's not there yet on the hardware level. But I think within a, within a reasonable, like single digit year horizon.~ Uh, ~you will see an incredible breakthroughs,~ um,~ in that area. We are also working on quantum computing to some extent in a kind of PSC setting. We just published [00:14:00] by nature, nature by Technology paper.
Wow, amazing. Using real quantum comput to generate a molecule. ~Um, ~so I think,~ uh,~ the tech automation and new technologies,~ um,~ related to quantum and maybe domain specific foundational models. We're developing, we call them large,~ uh,~ language life models,~ um,~ would be making difference,~ um,~ in a very short time period.
Amazing. Wow. Well, Dr. Alex, that's all we've got time for today, but really thank you so much for stopping in and having a conversation with us. Thank you so much.