Dr. Aziz Abu-Saleh
Member
The Knowledge Cloud.
To support our Chemistry to Cell initiative, we are building a biomedical knowledge cloud that fundamentally transforms therapeutic target discovery.
At its core, the platform unifies cell foundation models, chemical compound data, and biological context into a terabyte-scale, multi-layer large language model. It can work across multiple biological and chemical data types, including proteins, antibodies, small molecules, and gene-expression profiles, by representing them in a unified sequence-based format. The model supports both predictive tasks, such as drug-target binding, toxicity, cancer-drug response, and protein-protein interaction prediction. The result is a single navigable system where the relationships that determine therapeutic outcome are for the first time unified.
Unifying Biology.
Drug discovery is shifting from finding molecules that bind to finding molecules that work in the cells of patients.
Recent advances have turned protein-ligand structure prediction into infrastructure. A molecule that docks perfectly is now achievable on a workstation. Yet roughly 90% of clinical-trial candidates still fail, and the largest single cause, about half of all failures, is lack of efficacy, not bad chemistry. Patients do not get better even when the molecule binds the target with textbook precision.
The reason is structural. Binding answers whether a key fits a lock. Efficacy answers whether the cell, the tissue, and ultimately the patient respond to that lock being engaged. These are different scientific questions. The field has spent a decade pouring resources into the first while leaving the second largely unaddressed.
That gap is the next prize. The technical ingredients to capture it have arrived in the last 24 months. Open-weight cell foundation models, now part of our knowledge cloud, are what will carry drug discovery to the next frontier of efficacy.
The Disease–Protein Signal.
The central structure of the graph focuses on disease–protein associations, enriched by gene expression patterns, pathway context, genetic evidence, and druggability signals. These associations are encoded as weighted graph edges, allowing the system to capture not only what is connected, but how strongly and in what biological context.
AI That Navigates Biology, Not Just Data.
On top of the knowledge graph, we deploy AI and graph-based machine learning models that can traverse, reason over, and learn from the structure of the graph itself.
This allows the platform to surface targets that have never been identified by traditional workflows, not because the data didn’t exist, but because no human or linear algorithm could realistically integrate it at this scale.
By transforming raw biomedical data into a navigable, AI-ready knowledge graph, we are building the foundation for a new paradigm in drug discovery, one where machines can reason across biology to uncover therapeutic opportunities humans could not find alone.
From this foundation, our AI native systems construct custom chemical models tailored to each target’s unique biological constraints enabling the design of novel molecules at unprecedented scale and specificity.
