SyntheticGestalt Launches Molecular AI Models 'ZAO' and 'KOYA'

SyntheticGestalt Inc. (Headquarters: Shinjuku-ku, Tokyo; CEO: Yukiteru Shimada; hereinafter "the company"), a developer of molecular AI, announces the launch of two molecular AI foundation models: the molecular foundation model 'ZAO' and the molecular generation model 'KOYA'. These products support early-stage drug discovery by enabling AI to identify promising molecules from vast compound libraries and design new molecules, accelerating the cycle from exploration to optimization. Additionally, both models will be accessible via the 'NVIDIA BioNeMo Agent Toolkit' announced by NVIDIA in June 2026, enabling integration into AI agent-driven drug discovery workflows.

ZAO is a molecular foundation model specialized in small-molecule information, converting the chemical structures of small-molecule compounds into data that AI can process.

- Utilizes a '4D' representation that captures multiple 3D conformations of molecules, enabling the model to represent molecular shape and flexibility beyond what traditional methods (limited to 2D or single 3D structures) can capture - Pre-trained on approximately 10 billion compounds, allowing high-accuracy predictive models to be built even with only hundreds of in-house data points - Outputs stable, reproducible features regardless of the chosen 3D conformation for the same molecule - Achieved top rankings in 9 out of 9 public benchmark tasks—including ADMET and activity prediction—without task-specific fine-tuning, establishing world-leading performance as a standalone general-purpose model*

KOYA is a molecular generation model that designs novel, high-activity ligands for target proteins while precisely reflecting medicinal chemists’ design intentions.

- Enables precise control by fixing promising scaffolds and growing molecules only at specified positions and directions, directly applicable to lead optimization and fragment-based drug discovery - Generates novel molecules while preserving desired interactions with target proteins, with customizable constraints on allowable elements and molecular size based on chemist input - Automatically excludes synthetically challenging or unrealistic structures, proposing candidates that are drug-like and ligand-efficient, ready for immediate evaluation - Generated novel molecules outperforming known binding ligands in docking evaluations across 8 out of 10 evaluated targets

The NVIDIA BioNeMo Agent Toolkit provides a platform where functions such as structure prediction, molecular generation, and docking are standardized as skills callable by AI agents. By making ZAO and KOYA accessible through this toolkit, the company enables researchers to integrate molecular digitization (ZAO) and novel ligand design (KOYA) into unified workflows via AI agents, alongside other life science tools.

The company has previously demonstrated the effectiveness of its molecular AI foundation technology through large-scale compound exploration, including a joint research project with Enamine, a leading compound library provider, conducting virtual screening of approximately 38 billion compounds. It is also advancing technical validation with major domestic pharmaceutical and chemical companies. ZAO and KOYA represent the commercialization of the foundational technologies developed through these achievements.

Yukiteru Shimada, CEO of SyntheticGestalt, stated: "The era has begun where AI agents autonomously advance research. By making ZAO and KOYA accessible via the NVIDIA BioNeMo Agent Toolkit, agents can take on tasks from molecular evaluation to design, collaborating with researchers to accelerate discovery. We aim to realize this new paradigm of drug discovery."

Both ZAO and KOYA are offered to corporate clients. Usage models and pricing are available upon request for both products.

About SyntheticGestalt

SyntheticGestalt is a Tokyo-based molecular AI company developing and deploying AI for industries including pharmaceuticals, chemicals, materials, agrochemicals, cosmetics, and food. Since its founding in 2018, the company has aimed to build a system that mass-produces inventions through AI. In past research projects, it has discovered lead compounds for pharmaceuticals and low-environmental-impact materials, reducing the cost and duration of exploration by up to 90% compared to conventional methods. For more information, visit https://www.syntheticgestalt.com.

Inquiries

SyntheticGestalt Inc.

Minakami Aoi, President's Office s.minakami@syntheticgestalt.com

* Based on achieving first place on public leaderboards as a standalone general-purpose model in 9 out of 23 datasets from the Therapeutics Data Commons (TDC), including the ADMET Benchmark Group and DTI Domain Generalization Group (as of July 2026)