Nanyang Biologics' Vecura platform, powered by NVIDIA GPU infrastructure, is rewriting the rules of natural ingredient discovery.
The wellness industry has a dirty secret: most of the "natural" ingredients in your stress-relief supplements and focus-enhancing drinks come from a remarkably small, decades-old playbook. Caffeine. Theanine. GABA analogs. These legacy compounds have been recycled across thousands of products with limited mechanistic clarity and even less differentiation.
Meanwhile, the World Health Organization reports that more than 970 million people worldwide live with mental disorders, and post-pandemic demand for mood-regulating and stress-relieving products has never been higher. The gap between what consumers need and what science has delivered is widening.
Nanyang Biologics (NYB), a Singapore-based AI-native biotechnology company spun out of Nanyang Technological University, set out to close that gap; not with incremental improvements, but with a fundamentally different approach to discovery.
Project X: From 700,000 Compounds to Market-Ready Leads
When a global consumer health conglomerate approached NYB with a challenge, discover plant-derived compounds for emotional balance, focus, and stress resilience, the team turned to their proprietary Vecura platform and its specialized DTIGN engine for drug-target interaction prediction.
The results were striking.
Starting from a natural compound library of over 700,000 molecules sourced from Southeast Asia's diverse ecosystems — plants, fungi, animals, and insects — the team identified 3,500 compounds with high bioactivity and low toxicity across 156 key neuromodulatory receptors and transporters. These targets span the serotonergic, dopaminergic, glutamatergic, and GABAergic pathways most commonly associated with mood, cognition, stress response, and neuroplasticity.
The entire journey from target definition to lead recommendation took five months. By comparison, conventional high-throughput screening (HTS) typically yields hit rates below 1% — often as low as 0.01–0.1% — requires multi-day campaigns to process more than 100,000 samples, and costs $0.10–$1.50 per compound screened.
NYB's AI-driven pipeline delivered a 64× higher hit rate, 7× broader compound coverage than known databases, and 6× faster discovery and validation cycles than traditional methods.
The Engine Behind the Results: Vecura and LigoSPACE
At the heart of NYB's capability is Vecura, a modular discovery platform that integrates over 200 AI models into a single, no-code interface spanning the full small molecule discovery pipeline: from target identification and hit screening to lead optimization and preclinical validation design.
What makes Vecura distinctive is not just breadth but depth. The platform's proprietary AI engine, DTIGN (now evolved into LigoSPACE, or DTIGN 2.0), uses interaction-based 3D graph neural networks to predict compound-target interactions with a level of spatial awareness that most competing approaches miss entirely.
Where conventional models rely on ligand-based 2D representations, LigoSPACE models the spatial emptiness around ligands; the vacant space in binding pockets that determines whether a drug can successfully inhibit a target or will be displaced by endogenous substrates. It integrates multiple binding pockets through a unified geometric representation, combining techniques called GeoREC and Union-Pocket to capture protein-ligand interaction geometry with high fidelity.
The numbers speak for themselves: LigoSPACE delivers an 18.39% improvement in bioactivity prediction (Pearson correlation), a 9.93% improvement in accuracy (RMSE), and a 19.09% improvement in ligand ranking (Kendall tau) over baseline models. The work has been published in IEEE and presented at NeurIPS; rigorous peer-reviewed validation that sets NYB apart from platforms making claims without published benchmarks.