ZeroShotAPI Product Launch

Transfer learning works best for teams with years of well-structured historical data. Most teams don’t have it.

MemoryBO, our transfer learning algorithm, warm-starts new campaigns from related prior data, so optimization begins with what’s already known. But it has only ever been as strong as the data a team could bring to it. A biotech running its first process campaign, a CDMO starting a new client project, or a team whose historic results sit in slide decks and spreadsheets has been left starting from zero.

ZeroShotAPI changes that.

MemoryBO can now learn directly from proprietary data generated in the ReactWise HTE lab, starting with Suzuki and amide couplings. These are two of the most widely run reactions in pharmaceutical development. The datasets span a broad range of catalysts, ligands, bases, solvents and conditions, and include the negative results that most datasets leave out.

Filter the data to match your chemistry, and your campaign starts from what we have already learned. The first recommendations are informed by thousands of related couplings rather than intuition or a space-filling design.

For teams that do have their own historical data, the two combine. For teams that don’t, there is now a meaningful starting point either way.

This builds on our Innovate UK-funded ZeroShotAPI project with the University of Nottingham.