Announcements, talks and pharmaceutical AI insights — and the peer-reviewed work behind the method.
Every figure is from a completed project, traceable to a publication or a named partner.
Selective Cleaning reduced model error from 0.74 to 0.58 against standard preparation on the published MDM2 benchmark.
Coefficient of determination reached on the same MDM2 benchmark, the highest reported for that target in the published literature.
Classifier performance in the internal norovirus GII.4 peptide-design programme; this range is AUC-ROC, not accuracy.
One percent of the enumerated peptide library was docked to create labels before the model scored the full 603,750-member library.
Selective Cleaning weights each record, so conflicting lab, assay and instrument data stops degrading your model. Proprietary, and built for chemistry and biology — not adapted from generic ETL.
We have spent the last decade obsessed with the "brain": the model. The real barrier to AI discovery sits upstream, in the data.
Read the article →The pharmaceutical industry is currently navigating its most significant transformation since the dawn of the antibiotic era.
Read the article →The pharmaceutical industry has long been haunted by Eroom’s Law: while computing gets cheaper every year, drug discovery keeps getting slower and more expensive.
Read the article →We speak across the Southeast Asian pharma and biotech scene. Tell us what you're working on.
Bring your challenge.
Let’s find the right next step.