Projects and apps we have built for our clients.

DrugRev is a molecular engine we developed to explore the chemical binding activity, pharmacokinetics and side effects of drug-like molecules. It draws on a range of mined public and licensed sources — PDB, DrugBank, ChEMBL and FAERS among them — alongside whatever in-house data a client brings, with 3D visualization and docking on one side and statistical and deep learning models on the other.

Together these have helped our clients predict whether a molecule will engage its intended target in a clean and efficacious fashion. DrugRev also offers generative capabilities: designing small molecules and de novo binders — macrocycles, peptides and mini-proteins — with a specified activity profile against a particular set of targets, including the option to ensure the result does not engage designated off-targets.

Multiple datasets spanning various data types — GWAS, transcriptomics, proteomics, binary screens and other specialized measurements — are common in genomics, and integrating them is challenging. The process often gets reduced to a rigid funnel where only genes with positive results in every dataset are considered.

Bayesic selects an appropriate prior for each data type and converts each contribution into a weighted posterior probability. This enables a more flexible evaluation, where a negative result in one dataset does not automatically disqualify a gene, and you can see exactly how much each dataset moved it.

Target identification and validation is the request we get most often. Targette is the tool we built for it: GWAS, transcriptomics, single-cell data and pathway analysis in one place, arranged so that the chain of evidence from a variant to a phenotype stays visible.

Clients use it to go from a genome-wide scan to a handful of candidates that hold up across indications, then to check what else those candidates sit on — the ancillary pathways and protein–protein interactions that turn a clean target into an awkward one.

Every research organization has more written down than anyone can hold in their head: assay results, protocols, internal reports, and the literature around all of it. Grimoire is the assistant we build over that corpus — retrieval-augmented, running on a client-specific cloud deployment, and inheriting the same data access privileges as the account asking the question.

The design constraint that matters is provenance. Every answer names the chunks it came from, so a claim can be traced back to the report it came out of rather than taken on trust.