our mission_
a new foundation_
We’re building a truly foundational model for molecular biology and drug discovery. Not just better predictions. Not just faster drug design. A fundamental approach to understanding how molecules interact in biological systems. One that could transform every stage of drug discovery.
The Problem_
Today, bringing a new medicine to patients costs more than $2 billion and takes 10-15 years. Nine out of ten clinical drug candidates fail. Many diseases remain untreated.
Why? Because drug discovery is built on uncertainty. Every critical question is a molecular interaction we can’t reliably predict.
Will this molecule bind to its target? (Protein-ligand interaction) Will it reach the target in the body? (Membrane permeation, transporters, protein binding) How will it be metabolized? (Enzyme interactions) Will it cause toxicity? (Unwanted binding and metabolic effects) What will it actually do in cells? (Protein network interactions)
The way a cell works and interacts, how organs function, how the entire biological system operates – drug discovery is molecular interactions all the way down.
Yet for decades, predicting these interactions has been considered computationally intractable. A full quantum-mechanical description of binding is computationally too expensive and impractical at scale. The data is too sparse or noisy. The chemical space which contains more possible drug-like molecules than atoms in our solar system is too vast to explore systematically.
Our vision_
We’re building a new foundation for how humanity discovers medicine – a truly foundational model of molecular biology that can predict and understand the fundamental interactions governing life itself.
In an era where „foundation model“ has become the default term for large pre-trained AI systems like LLMs, we deliberately say „truly foundational.“ The distinction matters. Foundation models learn statistical patterns from massive datasets. In molecular biology, this approach isn’t enough. You can’t simply scale up compute and data to predict novel molecular interactions, because the data simply does not exist nor can it be generated at pace. And molecular interactions are more than statistical patterns.
A truly foundational model must be grounded in first principles. It must predict how the fundamental entities in the human body interact with each other. It must generalize to uncharted territory. Such a model would be a true foundation for drug discovery.
Building such a model requires careful curation of existing data to understand its usefulness and limitations, deep expertise in molecular processes, and exceptional capabilities in both quantum physics and machine learning. Master the fundamentals with unprecedented precision, and complexity becomes tractable. From this foundation, we can build increasingly sophisticated models: predicting how drugs are absorbed and distributed, how they’re metabolized, where toxicity arises, and how molecules modulate cellular pathways to multi-scale biological systems.
Delivering real impact today_
This isn’t hypothetical. After 5 years of intensive research and development in collaboration with Helmholtz Munich, we’ve created a system that understands fundamental chemistry, predicts molecular interactions at unprecedented precision, and generalizes to unknown domains where other approaches struggle.
The result is a model that has learned the rules governing molecular binding, not just the patterns in existing data. This translates into three critical capabilities already delivering real-world impact today: predicting protein-ligand binding structures through molecular docking, designing entirely new therapeutic molecules from scratch, and systematically optimizing lead compounds based on atomic-level understanding of binding.
The technology works. Our docking approach achieves 83% success rate on PoseBusters and 95% accuracy on CASF-2016 scoring benchmark. More importantly, we’re the only AI-driven system demonstrating impressive generalization capabilities in the Run’s Poses and PoseX benchmarks, proving our models work beyond their training data. The molecules we generate are chemically diverse, readily synthesizable, and precisely tailored to their target proteins, covering vast regions of chemical space that traditional methods cannot reach. Our optimization process is guided by deep insight into protein binding behavior, enabling rational, efficient improvement of drug candidates.
This platform is already deployed with pharmaceutical industry partners, at leading research institutions, and powering our own internal drug discovery programs targeting devastating diseases.
The mission is bold, the goals ambitious. But we’re already delivering impact today.
