Drug discovery is among the slowest and most expensive pursuits in modern science. Bringing a single therapy to market can take over a decade and cost up to US$3 billion ($3.84 billion), with failure rates so high that tens of thousands of molecules are typically tested before one even reaches human trials.
Singapore biotech start-up Nanyang Biologics believes artificial intelligence (AI) can change that by predicting which natural compounds will work before costly lab tests begin.
“Our vision is to create a search engine for molecules. If anyone wants to find compounds for a disease, food product or cosmetic, they can come to us. We provide the data and models that simulate the human body, saving years of trial and error,” Giang Nguyen, its chief technology officer, tells The Edge Singapore.
From herb garden to AI lab
The company’s roots lie in a Nanyang Technological University (NTU) herb garden. Founder and chairman Professor Roland Ong observed that some cancer patients who consumed certain plants appeared to improve, sparking systematic research.
Traditional testing quickly revealed the scale of the challenge. Each plant may contain tens of thousands of distinct molecules, and evaluating just one compound could take months.
See also: From AI beds to remote ICUs, start-ups are plugging India’s health gaps
“[The conventional method undergoes] a lot of trial and error because we don’t know which molecule actually works. That’s why it takes us so many years to discover new compounds,” Nguyen says. Scaling such an approach across hundreds of species was simply impossible.
The breakthrough came through a collaboration between NTU’s drug discovery and computer science departments. “Why don’t we collect a lot of data on how human proteins interact with different compounds in the past?” Nguyen recalls the founding insight. “If we can train such a model, we can prioritise which compounds to test first, instead of trying randomly as now.”
The result was the Drug Target Interaction Graph Neural Network (DTIGN), which scans millions of compounds digitally, ranks them by predicted efficacy and toxicity, and sends only the most promising to the lab.
See also: Connecting ecosystems: Build blocks, not billionaires
Search engine for molecules
Nguyen describes DTIGN as “a search engine for molecules”. Users can search for specific proteins in the human body and receive ranked results of compounds predicted to bind effectively.
For example, when someone searches for targets linked to dopamine or serotonin (relevant for mental health), the system returns natural compounds scored by predicted activity and toxicity. It also maps the sources of active compounds and provides visualisations of how they bind to proteins.
DTIGN does not take random guesses. It is trained on large datasets of past protein–compound interactions and uses graph neural networks and molecular docking to forecast the bioactivity of untested combinations. Results are refined continuously through feedback from lab experiments.
Beyond protein binding, the company aims to build a full physiological simulation or a virtual simulation of what happens inside the human body when certain molecules are consumed.
Scaling with Equinix
Running molecular simulations requires enormous computing power. “We need the infrastructure to continuously support the AI model to scan 24/7 and explore the vast chemical space,” says Nguyen.
To stay ahead of the latest tech trends, click here for DigitalEdge Section
As such, Nanyang Biologics relies on Nvidia’s graphics processing units (GPUs), HPE’s supercomputing servers, and Equinix’s data centres. “With all these pieces together, we can massively scale up the operation of compound discovery, which will eventually give valuable compounds that save lives,” he adds.
Privacy was another factor. Nguyen says: “Information about molecules and related data is proprietary and unique to us. So, we want to keep it secure in a private server while not compromising our ability to scale because Equinix has connectivity worldwide.”
Eric Hui, business development director for Internet of Things (IoT) ecosystems at Equinix Asia Pacific, shares that the partnership reflects broader industry trends toward private deployment for sensitive intellectual property. “About 30% of the volume of data in the world comes from life sciences. Those companies like Nanyang Biologics need scalability, privacy, security and global reach — all of which Equinix can offer with our AI-ready data centres worldwide.”
A data company
Nanyang Biologics does not plan to manufacture drugs. Instead, it charges customers for access to its compound database and predictive scores via DTIGN, while partnering with labs for verification and optimisation.
“We want to be the data company for compounds,” says Nguyen, adding that DTIGN’s unique edge lies in its vast library of bioactive molecules from nature in tropical regions. According to studies, such compounds are often less toxic than synthetics and feature complex structures that are difficult or impossible to replicate in the lab. However, those intricate structures allow unique interactions with biological systems. By screening these natural molecules at scale, DTIGN can accelerate discovery while grounding treatments in the proven potential of nature.
DTIGN is offered through flexible business models: subscription access, project-based contracts, milestone-linked partnerships, and co-development or licensing deals with revenue sharing.
A freemium version is being piloted for academic and early-stage users, with premium tiers unlocking advanced tools and IP support. “Each engagement is tailored to a client’s validation strategy, regulatory path and commercial goals,” Nguyen says.
That approach is gaining traction. The startup has grown from one collaboration last year to more than seven today, spanning pharmaceuticals, cosmetics, food and traditional Chinese medicine. They include French cosmetics firms exploring natural skin pigmentation modulators and Korean partners expressing interest in compounds for hair growth and removal.
Food applications could be even larger, given the physiological impact of natural compounds. “We all know that coffee has dual neurochemical effects: boosting energy and suppressing sleep hormones. But in the future, there might be a compound that can influence oxytocin, making you feel loved in a friendly, non-toxic way through diet,” exemplifies Nguyen.
Next milestones
Nanyang Biologics has offices in Singapore and Vietnam, with plans for Hong Kong and the US, targeting regions with substantial pharmaceutical research activity.
The company is also developing a second-generation platform that aims to provide comprehensive physiological modelling. “DTIGN 2 will be even more accurate than the first version. It will contain the knowledge of compound signatures and be able to predict a compound’s effects on the human body,” says Nguyen.
The path ahead, however, brings significant hurdles. He highlights that talent is scarce, especially people who straddle both biology and AI or computer science.
Bias in AI models presents another risk, as predictions are only as good as the training data. “You cannot leave the AI to do everything. Human experts must verify outputs, and lab results need to retrain the models,” asserts Nguyen.
Regulatory hurdles may also slow adoption as new approaches often fall outside existing frameworks. Nguyen has therefore called for governments to foster open innovation between startups, academia and industry.
Whether AI can crack drug discovery’s multibillion-dollar problem is unproven, but if Nanyang Biologics’ search engine delivers, it could show how digitising nature’s pharmacy can cut costs, shorten timelines and open new frontiers.
