5Y Talk | How Can Computing and AI Break Through the Traditional Paradigm of Drug Discovery?

Information technology is penetrating every field at an unprecedented pace, reshaping the future of every industry. Medicine — the industry most intimately connected to our health and daily lives — is facing a paradigm shift under IT's influence.

As one of the earliest investors to bet on the "ITBT" track, over the past few years we've discovered and supported many entrepreneurs in digital healthcare. They've thrown themselves into this field without hesitation, using technology to break through industry bottlenecks. From a longer time horizon, IT's penetration into medicine has only just begun — much more change lies ahead.

Where exactly do the pain points and challenges of traditional drug R&D lie? What transformations will technology bring, and what value will it create? How have the pioneers explored and practiced? Recently, 5Y Capital and VB Data jointly hosted the "Digital Pharmaceutical R&D (ITBT)" forum.

For the panel discussion, we invited founders from 5Y Capital portfolio companies — Yang Li, Chairman of ReviR Therapeutics; Chengtai Li, Founder & CEO of Galaxy Pharmaceutical; Caida Lai, Co-founder & CEO of METiS Pharmaceuticals; and Hang Chen, Co-founder & CEO of Neox Biotech — for a discussion on "How Can Computation and AI Break Through the Traditional Paradigm of New Drug Discovery?"

We've compiled portions of the panel, hoping it offers some insight :)

Yang Li: Please start with brief introductions, and why you chose to start companies in this field?

Chengtai Li: I'm Chengtai Li. Our company, Galaxy Pharmaceutical, was founded in 2019. We're dedicated to using AI to accelerate early-stage drug discovery, with a current focus on small molecules.

This grew out of my experience in Boston. My undergraduate and MIT background is purely in IT, mainly AI. In 2017, AI technology was making exciting advances, and Boston is a global biotech hub — so there were many interdisciplinary opportunities, and different companies emerging. Ours was born from that moment.

Caida Lai: I'm Caida Lai from METiS Pharmaceuticals. METiS was spun out from XtalPi in January last year, specifically to use AI for drug delivery. The company originated from my partner, Academician Chen, and myself connecting with XtalPi and realizing that the drug delivery formulation space had been largely ignored — yet it's foundational work requiring massive cross-disciplinary integration. So we started METiS.

Unlike molecules, drug delivery has many underlying, foundational high-throughput experimental platforms and algorithmic platforms that haven't been built yet. So over the past year we've focused on combining big data platforms, quantum computational algorithms, and AI into simulation platforms. Once these were established, we began working with partners to advance clinically differentiated improved new drugs.

Hang Chen: Hello everyone, I'm Hang Chen, Founder and CEO of Neox Biotech. Neox was founded in 2018, now based in Beijing's Zhongguancun and Shanghai's Zhangjiang, and we're a resident company at JLABS @ Shanghai by Johnson & Johnson. We're a next-generation computationally designed pharmaceutical company, focused on large-molecule and multispecific drug R&D. In large-molecule drug discovery, we mainly combine AI with biophysics. We've also seen the rise of biotechnologies like synthetic biology that can generate data at scale. Combining these three — AI, biophysics, and high-throughput experimentation — we believe this is a fundamentally new methodology for solving cutting-edge problems.

You could understand our company this way: 30 years ago there was a company called Vertex. They believed that understanding structure could help make drugs, and they were right. Five years ago, many companies realized that understanding structure alone wasn't enough — molecules are constantly moving, so you need to study dynamics and simulation. We believe molecular motion is just a monomer problem; we study protein-protein interactions. The monomer problem is already complex, but protein-protein interactions are even more so. We want to do drug R&D based on understanding complex systems, hoping to make innovative drugs from this angle.

Yang Li: Thank you all. I'm Yang Li from ReviR Therapeutics. We focus on RNA computation. One end of the central dogma, DNA, was digitized long ago, with many sequencing companies continuing that work. At the other end, protein structures are increasingly understood, and protein sequencing is seeing startup activity. In the middle, RNA structure — new methods have emerged in recent years, while sequencing costs keep falling. We use algorithms and large-scale sequencing to digitize RNA structure, then use the most primitive, simplest modality — small molecules — to target pockets within RNA structures, attempting to solve many "undruggable" targets at the protein level. We're a startup incubated by Aegicare, a genomics-based disease innovation diagnosis and treatment company.

Your introductions were all quite thorough, and you're each切入 AI pharma from different angles. I'd like to ask: how do you yourselves define AI pharma?

Chengtai Li: As I mentioned, the pharmaceutical industry faces serious challenges — R&D costs are very high while efficiency is very low, and everyone's thinking about how to change this. At that time, AI had new breakthroughs, and people wondered if AI could provide value.

AI can indeed provide value, but it can't solve everything. Although we're an AI pharma company, AI can only solve part of the problem. Many issues require other methods — biological, physical, or high-throughput experimental approaches. Solutions are diverse; AI is just one. We treat AI as a tool; the real goal is solving problems.

Yang Li: Collaboration with pharmaceutical companies is very important for AI pharma. In your exchanges with pharma companies, what do they tend to focus on, and what insights can you share?

Chengtai Li: Domestic pharma companies differ from MNCs, and large versus small pharma companies have different needs. Some needs are relatively modular — for instance, how to find new hit compounds for new targets, or how to modify structures based on existing drugs to break through original patents. The problems they face vary.

Hang Chen: We have very deep collaborations with both MNCs and Chinese listed pharma companies. Their needs are diverse, but our main offering is original innovation. For example, how to screen better antibodies for certain pipelines. We also do more differentiated work — nanobodies, bispecific antibody design. We basically solve difficult problems and provide differentiated value.

Caida Lai: I mentioned two aspects earlier — whether we can collaborate at the pipeline level, and how to get clinically differentiated products to market faster. Let me add another point: you can think about this on two levels, one is efficiency, and the other is whether we can push the boundaries of expert knowledge beyond where they currently are.

Initially pharma companies wanted efficiency-focused collaborations. But more recently, we're seeing increasing collaborations around breaking through expert knowledge boundaries and expanding drug design space. They hope that through working with us, they can do many things they couldn't do before, completely breaking out of their initial design frameworks. This seems to be the direction they're more interested in now.

Chengtai Li: I strongly agree with what Caida said. We've encountered the same demand from many pharma companies — they care whether you can do what they can't do, what your differentiation is. This is the question we face.

Yang Li: I've heard that especially with overseas pharma companies, rather than telling them you'll improve early-stage speed by 50% or reduce certain costs by 50%, it's better to say you'll help reduce the probability of Phase III clinical trial failure by 10%. This relates to optimizing many algorithms — for example, in DMPK, how can we incorporate considerations early on and come up with an optimal solution?

Chengtai Li: That's a great question. Actually, both parts of drug R&D — preclinical and clinical trials — have high failure rates: over 90% for preclinical, and over 90% for clinical as well. What we're solving now is more the preclinical side — PDC, IND filings. How to solve the clinical problem is fundamentally a biological question, requiring better biological models to evaluate whether something works in animals. This is something everyone needs to think about.

Yang Li: In interdisciplinary fields like AI and computational pharma, what kind of talent do you look for? For example, do you need people with computational backgrounds to learn biology, or biology backgrounds to learn computation? Can you share your hiring experience?

Caida Lai: The ideal person understands both sides — quantum computation and AI, plus biology and medicine. We've been trying to find this person, but I currently feel they probably don't exist. So the key question is: how do people with computational thinking, people with traditional experimental thinking, and people with decades of pharma experience communicate with each other? From initial team building to current culture building, this has been an important issue.

When I was at MIT, there was a saying: computational people know everything but do nothing; experimental people do everything but know nothing. And these two sides basically don't communicate with each other. Now at our company, we're trying to build cross-disciplinary communication. When I design a drug delivery process, if we can't find results experimentally, computational people can analyze the most fundamental quantitatively-derived information. One person is unlikely to understand everything; what's more needed is good communication.

Hang Chen: These interdisciplinary people do exist — they just went off to start companies, like you.

Yang Li: Speaking of cross-disciplinary跨界, there's another trend now where many internet giants are entering AI pharma. What do you think? What signal is this for the industry?

Chengtai Li: It's not just internet giants paying attention — many pharma companies themselves are doing this too. AstraZeneca uses AI in more than half its pipeline. This is actually a very clear signal that everyone is starting to recognize the value of this.

We started our company in 2019. On one hand, the market had been educated to some degree; at the same time, this pandemic has made everyone realize the importance of this, and people are starting to feel that the entire field has room for improvement, that the efficiency problem needs to be solved, and AI can indeed provide certain value. The entire market is still a vast blue ocean. Everyone paying attention to this is actually a good thing — it can push the entire field forward more rapidly.

Caida Lai: Chengtai put it very well — it's about ecosystem building. When I was at McKinsey & Company in 2016-2017 advising pharma companies, European and American pharma companies had already started doing this, and they were collaborating with Microsoft and Google from the beginning. In fact, they'd already begun full-chain digital transformation. Domestic giants are just entering this field now; they'll be more like IT partners for pharma companies, while we want to be more like digital pharma companies — there's an essential difference in where we sit in the industry chain. The more players in this ecosystem, the more it can help the entire industry transform digitally and through AI.

Yang Li: Finally, everyone can speak freely. What did you imagine for your company or the industry's future in the early days? Looking at today, how much do you feel has been realized? What stage is the industry at now?

Hang Chen: I recently read a book called Think Like a Rocket Scientist — very interesting, and I've recommended it at our company. The book mainly talks about the American moon landing — at the time, scientific progress wasn't sufficient to support it, but they accomplished it anyway. What impressed me most was the relationship between how you define problems and the tools at hand. People like to use the tools they have to define problems, and of course, how to define a good problem is very important. On the other hand, tools themselves have advantages, and you should fully utilize the characteristics of the tools you have.

So I've been thinking about the advantages of AI pharma. Solving one drug creates value, but can't traditional methods also solve one drug? Traditional methods can also solve two, three, more drugs — it's all additive in principle.

So where does AI pharma's greater value lie? I don't think this is an additive principle matter. For example, in compound screening: traditional companies have a pool, they fish out one "fish," and that fish is gone. Fish out more, and the pool depletes. But we use data-driven approaches to learn from the fished-out fish, generating new fish, bigger fish — the fish in this pool actually increase. Theoretically, the barrier should get higher and higher. This is our expectation. I also remind myself whether what we're doing is actually like this.

Second, can AI enable personalization? Personalization is essentially what AI or data-driven approaches can fundamentally change — personalized medication, and later possibly personalized payment? Our expectation for ourselves is not just to make one drug. Boldly imagining, we also hope to eventually do personalized payment, even personalized treatment hospitals. Cell therapy, gene therapy are indeed personalized; perhaps we can really complete the infrastructure construction to realize these personalized things. This is very exciting for us.

Caida Lai: AI pharma is currently in a phase of bold hypotheses, but still careful verification. It's a bit like the internet industry in the 1990s — some infrastructure had been built, maybe hundreds of internet companies would rise, but most would fail. But great companies would certainly emerge.

In the careful verification process, what's important? How do you verify a long-term viable business model? It requires very long-term capital investment, continuously accumulating differentiated capabilities, accumulating underlying barriers, to become a great company.

Second is the translation from preclinical to clinical development — most companies currently don't have this capability, and this is a crucial part for truly making this track work in the next step. So I think going forward, the industry will experience rapid growth, there may be some crashes, and then several truly great digital pharma companies will emerge. This is the direction for the next 10 years.

Chengtai Li: I very much agree with what the two just said. At the same time, I'd like to add that I believe the entire industry is currently at a relatively early stage. What counts as ultimate success — business model is very important on one hand, but true success depends on whether you can actually deliver, and whether what you deliver is useful to patients. This is something we care deeply about.

I feel like we're running a marathon, seeing who can break two hours. Now some foreign companies have just broken two hours. Who's next? The entire industry is now on a rising trend, and I suspect we may also encounter a crash, after which things will return to rationality — but we're already seeing the dawn. That's my thinking.

Yang Li: Picking up on what Hang Chen said — internally, we never say we're doing science; we say we're doing engineering. The logic of engineering is that regardless of success or failure, each completed pipeline accumulates experience, and the next pipeline has higher success probability than the first, rather than each time being a eureka moment, relying on flashes of insight with no way to predict the next success probability. I believe we're all working in this direction.

5Y Capital (formerly Morningside Venture Capital) currently manages approximately $5 billion in USD and RMB dual-currency funds. We believe that if the "crazy" you in others' eyes begins to be believed in, the world will be a better place.

BEIJING · SHANGHAI · SHENZHEN · HONG KONG

WWW.5YCAP.COM