Traditional Bayesian optimisation starts from zero every time. MemoryBO doesn’t.

Traditional Bayesian optimisation starts from zero every time. MemoryBO doesn’t.

In our last post, we explored the value of data-driven machine learning for chemical process optimisation. But there’s one advantage that chemical intuition - and the scientific literature - still holds over standard machine learning models: it learns from experience.

Chemists learn from what worked and what didn’t. This guides future campaigns, meaning that redundant parameter space can be ignored, and good practice continued. 

Unfortunately, those learnings are often inconsistently recorded and poorly communicated, or never written down - and with thousands of data points from large labs, it's impossible to keep track of them entirely. 

MemoryBO changes that.

MemoryBO is our proprietary multi-task Bayesian optimisation engine designed to transfer knowledge between related systems and intelligently warm-start new optimisations. 

It’s powered by thousands of high-quality data points spanning different chemical transformations and process conditions, generated in our HTE lab - focusing on the key reactions used in modern-day medicinal chemistry, including Buchwald-Hartwig-, Suzuki-, and amide couplings.

By reusing prior knowledge to guide exploration, it delivers faster convergence, smarter experimentation, and earlier identification of high-yield regions that standard BO would take much longer to uncover. The broad space covered by our datasets, and the inclusion of negative results, allows you to focus on experiments that really matter. 

Even if you have very specific chemical transformations that are not covered by our proprietary dataset, you can simply upload an Excel/csv file of the data from a previous campaign. We structure it, train a model, and help you unlock years of stored data.

Don’t let your learnings be forgotten. Unlock them with MemoryBO!