Screen in silico
Validate in vitro

Graph-based virtual screening of 60B+ feature complete hypotheses—Target, Mechanism of Action (MoA), Cell Type, & Inhibitor—for rapid, parallel, and unbiased testing of hypothesis space.

Picking a single MoA to screen against is risky business.

We developed a platform that creates, simulates, and ranks >60B feature complete hypotheses, flipping these risks into advantages. Testing diverse, targeted, and novel hypotheses, in parallel, maximizes success and speed.

Why Graphs?

By convoluting directed acyclic graphs at multiple ontology levels, we gain three major advantages:

MoA Scaling—Leveraging the topology of the graph itself, we can identify upstream, causal ontologies that drive a given transformation.

Gene Profiling—Each in silico gene knockdown is mapped across thousands of ontologies, allowing for predictions of which processes would be affected by a gene perturbation—resistance mechanisms included.

Hypothesis Filtering—By harmonizing our graph with additional metadata, we can filter for specific ontologies, targets, or transformations that meet the specific needs of a lab (e.g. Novelty? Existing probe? Expressed in specific tissue?)

Bayesian driven FEP for hit/lead discovery

Using Bayesian-guided Free Energy Perturbation (FEP) methods, we efficiently explore chemical space for compounds that meet specific requirements (e.g. solubility, potency, novelty, etc). This process can either search existing libraries or use synthesis-aware enumeration of fragments to generate novel compounds.

By leveraging structure-based pharmacophores, we can quickly create enriched libraries that can be evaluated in an active learning protocol that can reach lead-like potencies in ~weeks.