Machine-learning tool mines lab notebooks for lessons from failed reactions

A team of researchers has created a machine-learning tool to analyze failed chemical reactions. The system is designed to sift through misfires in researchers’ lab notebooks in search of insights that can support more accurate predictions about what reactions are likely to work in the future.

Typically, there is a disparity between the influence of successful and failed reactions. While details of what works spread through publication in research literature, the lessons that can be learnt from failed reactions remain hidden in the lab notebooks of individual researchers. Now, a team from Haverford College has set about trying to develop a tool that can guide material synthesis choices by unlocking these insights.

“We used information on ‘dark’ reactions--failed or unsuccessful hydrothermal syntheses--collected from archived laboratory notebooks from our laboratory, and added physicochemical property descriptions to the raw notebook information using cheminformatics techniques,” the researchers wrote in a letter to Nature. “We used the resulting data to train a machine-learning model to predict reaction success.” The machine-learning tool was designed to analyze these data and predict whether a set of reagents would yield a crystalline material when combined with a solvent and heated.

In the project reported in Nature, the tool looked specifically at templated vanadium selenites. When asked to recommend reactions within this niche, the algorithm proposed successful combinations and conditions 89% of the time. Researchers with a combined 10 years of experience with the materials were right 78% of the time.

Having generated evidence to suggest the algorithm can best human intuition, the team built a decision tree to guide their own decisions in the future. The tree allows researchers to work through questions such as "is sodium present?" to reach a prediction regarding the likelihood of a reaction resulting in a crystal.

In the longer term, the team hopes to see the project as a launchpad for greater use of data from failed experiments. The team is trying to facilitate this process by setting up the Dark Reactions Project, a scheme designed to encourage other researchers to share details of the own unsuccessful reactions.

“Failed reactions contain a vast amount of unreported and unextracted information,” Alex Norquist, a materials-synthesis researcher on the team that carried out the work, told Nature News. “There are far more failures than successes, but only the successes generally get published.”

- read the research report (PDF)
- and Nature Newstake