Bullfrog AI’s CEO Vin Singh on the TechBio boom

Artificial intelligence (AI) has rapidly become ubiquitous in biotechnology, fueling a steady stream of announcements about partnerships between biopharma companies and AI-focused firms. But who are these AI-driven companies, exactly? The answer lies in TechBio, a growing sector at the intersection of AI and biotech.

Indeed, the global market of AI in biotech was valued at $2.73 billion in 2023 and is projected to grow at an impressive compound annual growth rate (CAGR) of 19% to reach $7.75 billion by 2029. The number of AI-driven collaborations in this space is also surging – from 10 collaborations in 2015 to 105 by 2021. This trend is no longer based on speculations as AI-discovered drugs have shown success rates between 80% to 90% in phase 1 trials versus the historical industry average of 40% to 65%. 

To discuss the TechBio movement, we spoke with Vin Singh, founder and chief executive officer (CEO) of Bullfrog AI, one of the TechBios gaining traction by integrating its proprietary AI into pharma and biotech workflows. In this interview, he shares insights on the rise of TechBios, the challenges of AI adoption in biotech, and where the industry is headed.

The rise of TechBios in biotechnology: How do they work?

Many AI-driven companies are being labeled as TechBios. What defines a TechBio company?

Biotechs lead with biology, and TechBios lead with technology. TechBios are not interested in developing drugs – that’s what biotechs do. We don’t want to raise hundreds of millions of dollars and spend many, many years to develop a drug. Instead, we focus on using our technology to accelerate, advance, reduce investment, and increase the odds of success in some stage of development.

First, I think TechBio companies should have proprietary AI. They’re not just AI application companies, using it as a tool like any other company. 

The second thing is the specificity, TechBios are not generalists. They address certain problems that have been identified by the pharma or the biotech they work with by leveraging the available data.

TechBios operate as collaborators, rather than competitors to biotech firms. How does a collaboration between a TechBio and a biotech usually go?

We’re [Bullfrog AI] taking a collaborator’s information or assets, and helping them be more successful. That’s our responsibility and we’re doing this a couple of different ways. 

The first thing is target discovery, and we have access to a large repository of brains for neuropsychiatric disorders. We’ve analyzed all that data, we’ve discovered a number of drug targets that we call causal drivers of schizophrenia, bipolarity, and depression.

The other part of the business is what we call technology solutions. We know there’s a demand from biotechs to have an AI partner but they don’t have millions of dollars to inject into it. 

So, we thought, what if we can bring our technology to them in a sort of subscription model? We can build networks using our proprietary technology, out-of-public data, and proprietary data. We then combine them, and this enables these companies to advance faster and more successfully get to that next funding milestone to continue on their drug development path. 

What is the business model for TechBios? 

One is typically like a fee-for-service type of model and we aren’t currently asking for any intellectual property (IP) rights, although that could change in the future. I mean, if we’re making billion-dollar discoveries for companies, that’s worth a lot more than, you know, a few $100,000 right? 

On the other side, the structure of our target discovery business deals is pretty well established. Typically, there’s an upfront payment and milestone payments associated with the successful development, starting from a target.

So in both of these scenarios, we’re not competing at all, we’re partners. We’re trying to inject value into the marketplace.

How do TechBios’ solutions integrate into a biotech’s workflow?

TechBio companies can be integrated at various stages of drug development, but the biggest challenge is data availability and organization.

What data does the partner have? How is it organized? Do they have enough data? Do they have the right data? The data we’re most interested in are omics because they allow us to uncover the most insights. Also, the data may not be centralized, it may be spread out all over the world, even. 

The good news is that the cost of generating high-quality data has come down. So I think this will become more routine now. 

What are the specific challenges in collaborating with pharma or biotechs for TechBios?

Pharma is an established market for AI partnerships, they have the money and they know exactly what they want. Also, they’ve already extracted value from these partnerships. They’re the perfect partner. 

Biotechs are a lot more challenging because they don’t necessarily have the same budget and they don’t really know what they want. You’ll hear comments like: “Oh, run it in your AI and see what comes out.” And that’s not really a project. We have to have some clear goals for the project, and clear deliverables, to add value to the program. 

However, I’m starting to see a shift now. I think investors, and venture capitalists (VCs), are going to start putting definite budgets toward AI tools as part of their investment. Biotech is such a high-risk business, you want to do something to improve your chances of success. The investment in AI would be nominal compared to the cost of discovering and developing these drugs. 

What should a biotech pay attention to when looking for a TechBio to partner with?

I would advise to look at where their core technology comes from. If they are applying existing tools, maybe you want to put those to the side. If it’s proprietary, and has been validated in different industries, different sectors, that means it’s a very versatile technology. 

The other thing to consider is whether the technology can work with any type of data. Are the data you have and the capabilities of the AI platform a good fit? 

TechBios and AI in biotech: Hype or real game changer? 

What’s driving this trend? Is it funding, technology advances, or industry demand?

I think it’s a mix of all of those things. There’s a desire to fundamentally change the development paradigm of biotech. It’s a very high-risk business and there’s so much excitement: Big tech companies are encroaching into life sciences, tools such as OpenAI’s are more and more present. So, it’s definitely a combination of factors all coming together. 

I also think we’ve reached a comfort level with AI, because it’s in the news, and it’s being adopted across many different industries at this point. I think that familiarity helps a lot and it’s going to benefit everybody in the long run.

AI has generated a lot of buzz. Where do you think it is truly making a difference? 

It’s definitely making an impact on drug target discovery, and I think it’s going to lead to more successful targets. Most drugs fail because the wrong target is picked. And with technologies such as ours, we’re able to explain and illustrate how the identified targets have the highest probability of success.

And then, on the other end of the spectrum, AI will help with clinical trial design. I think the beginning stage, though, is the most important – you don’t want to spend hundreds of millions of dollars in 10 to 15 years, and realize you have the wrong target, right? 

AI is highly reliant on data to be efficient. Could you say more about how that translates in biotech-focused AI models? 

At Bullfrog AI, we use something called unsupervised machine learning. Unsupervised machine learning can find relationships that exist in data without any training. You can supplement that with public databases. 

When the training aspect comes into play, the tricky part about our industry, compared to every other industry, is the small sample number. For example, in a clinical trial, you might have 40 patients, whereas in every other industry, the sample numbers are millions or billions. That can be a challenge. 

Your technology has to be able to operate with a small sample number, and a lot of variables. 

What’s next for TechBio?

You said TechBios weren’t interested in developing drugs. Do you see that changing and TechBios eventually moving into developing their own therapies? Do you see a future where TechBios would merge with biotechs to create hybrid models?

Yes, both. Our plan right now is to focus on target discovery, get deals done, get the cash flowing, and then look to extend further out in the discovery process. But we’re going to do it in a step-wise fashion, as we build the company and we’re more and more successful, we will stretch ourselves out to a point where we may look more like a biotech company. Will we go as far as clinical trials, which is really where the money is spent? We’ll see. 

But I think overall, what you described is already happening. If you look at the merger between Recursion and Excientia – they’re essentially becoming a biotech company. Schrodinger is now trying to use its capabilities to design its own drugs and advance them. 

If you’re a TechBio company and your technology has proven itself, it’s a smart decision to extend further down. The further you take the drug, even in the discovery stages, the more the risk goes down and the value goes up, and you can do bigger deals. 

What does the next decade look like for TechBio? 

I underestimated how fast these large language models (LLMs) were going to develop. We’re already at a point – and I thought this was going to be years from now – where you have a bioinformatics AI agent.

I think 10 years from now, you’ll have one person, an army of AI agents, and you will tell the agents what you want done and it will get it done. Obviously, in drug development, there are still things that you have to do in the lab. We still have to do animal studies, and we still have to do human clinical studies. 

I don’t think any of these requirements for in-lab proofs will change in the next 10 years. Maybe beyond that, I can potentially see things like digital twins happen and being relied on. A whole-body digital twin is many years away, but an organ-level digital twin, though, I can see that happening in the next 10 years for sure. That’s tied to generating data but as it’s become more affordable, we’re going to have more high-quality data, which is what you need to create a digital twin. 

It’s going to be an exciting time, but I can’t wrap my head around it completely, because it’s moving so fast.