What if you could cut your experimental effort by up to 80%?

What if you could optimize your reactions without Design Of Experiments, cutting your experimental effort by up to 80%? Would you do it?

For years, Design of Experiments (DoE) has been the trusted method for optimizing chemical reactions in process research.

It’s a well-established, systematic approach, but it can be time- and resource-intensive, especially as the number of variables grows.

Take a common 5-factor optimization: solvent, base, catalyst, time, and temperature.

A full factorial categorical DoE would require >50 experiments with fixed time & temperature.

Only then the continous variables would be optimized in a secondary screen, requiring tens of additional experiments.

While comprehensive, this approach tests many conditions that ultimately don’t contribute to finding the best result.Well, now we have a faster way.

Bayesian Optimization (BO) is an emerging approach that keeps improving with easier access to digitized chemical data, improved machine learning models, and more powerful computing.

Instead of exhaustive experimentation, BO learns from each experiment and adapts in real-time, balancing exploring and exploiting the space to focus on the most promising conditions.

In the same 5-factor example, BO can often achieve the same result in just <20 experiments, cutting experimental effort by up to 80%.

At Reactwise, we’ve even seen more impressive efficiency firsthand: For the optimization of a CH activation reaction, a DoE campaign would have required >750 experiments to reach the optimal result.

Using BO enhanced with knowledge transfer from a related campaign, we found the ideal conditions in just 23 experiments, saving 97% of experiments - less time in the lab, fewer resources used, and faster insights.

As reaction optimization evolves, Bayesian methods are becoming indispensable tools for modern process chemists—adaptive, efficient, and ready to handle the growing complexity of reaction spaces.