Open evaluation tools, research access and a drug-discovery prize

1. X-Chem wins the first TBXT Challenge prize¹

The Chordoma Foundation announced on 14 September that X-Chem had earned the first $100,000 TBXT Challenge prize. X-Chem combined DNA-encoded-library screening with medicinal chemistry, computational analysis and machine learning to prioritise compounds. The Foundation says its laboratory repeatedly measured one compound binding brachyury below the competition’s one-micromolar threshold, followed by an independent chemistry review. The result gives the challenge a concrete, experimentally checked milestone for a difficult cancer-associated target. It is an early starting point for drug development, and the announcement does not establish an effective treatment.¹

2. Twist adds laboratory-data services to Lilly TuneLab²

Twist Bioscience announced an agreement with Lilly TuneLab on 16 September to provide antibody-characterisation data services, including support for the AbLab developability prediction model. According to Twist, participating users will be able to order laboratory services using preferred protocols and pricing, then contribute resulting data to TuneLab’s federated training. The arrangement connects model-based candidate selection with experimental measurements, a practical consideration for teams whose antibody designs need testing. Its significance lies in that proposed route between predictions and standardised laboratory work; faster discovery remains an expected benefit rather than an independently demonstrated outcome.²

3. UKRI announces a £5 million AI-adoption funding call³

UKRI has announced a £5 million opportunity for collaborations between AI specialists and domain researchers, including teams in advanced materials, engineering biology and medical research. Published on 17 September, the call opens on 29 September and closes on 1 December. Eligible UK research organisations may propose projects costing up to £250,000, with UKRI funding 80% of the full economic cost. Projects must last no more than a year and start on 26 April 2027. The call offers a concrete route to test AI research workflows, although funding remains subject to Full Business Case approval.³

4. OpenADMET releases tools for blind molecular-prediction challenges⁴

OpenADMET announced an open-source blind-challenge template on 16 September, combining scoring infrastructure with a participant interface. The developers describe support for regression, classification and protein–ligand pose prediction, with bootstrap-based uncertainty estimates and separation of hidden test labels from the public interface. A companion tool checks chemical train–test overlap, near-duplicates and identifier problems. Both are offered under Apache 2.0 licences. These resources give benchmark organisers reusable machinery for testing molecular models against withheld measurements. The announced features have not been independently deployment-tested here; releasing infrastructure does not by itself establish the fairness of any particular benchmark.⁴

5. OpenADMET plans open transporter data and blind benchmarks⁵

OpenADMET announced a partnership with the OpenAI Foundation’s Public Data for Health programme on 15 September to build open experimental resources for predicting compound distribution. According to the consortium, Octant will profile physicochemical properties and transporter interactions across tens of thousands of small molecules, while UCSF researchers will determine transporter structures. OpenADMET plans to release the resulting datasets and run prospective community blind challenges. This connects named experimental teams with a public testing route for models of transport liabilities. The announcement describes work to come; it does not report completed datasets or validated predictions of clinical brain exposure.⁵