In the past few years, machine learning topics have been on the rise as its applications spread to new areas of science, revolutionizing scientists’ perspective on all parts of the scientific process through the interpretation and handling of big data. Big data is a blanket term for the feature-dense collection of data that are unique to different facets, like historical pricing of products, patient data, linguistic corpora, and more. It can often take analysts loads of unnecessary time surfing through these data for correlations, trends, and outliers that might be notable in exploring further or offer a new outlet for discovery. Here’s where machine learning comes in: it utilizes quantitative features of the data, including probabilities and other trends, to make useful predictions and other discoveries that researchers can use to focus their innovation.
One interesting and recent venture utilizing machine learning is unfolding right now in a chemistry lab at Rice University. Catalysts are being improved through an analysis of elemental and binding properties of different metals involved in chemical processes. Catalysts help speed up chemical reaction rates through their different interactions with other reactants; controlling the actual interactions and any other possible byproducts are key to improving chemical processes and efficiency. Catalytic efficiency is paramount in a number of industries, including the automobile and pharmaceutical ones; it may even help reduce costs and environmental impact.
For researcher and professor Thomas Senftle, manually searching through a number of factors associated with the catalytic design was not efficient enough to keep up with the pace of academic research; however, analyzing these factors collectively was key to improving efficiency given changes could occur with different metals and even different kinds of metals. Some of the most interesting features included binding energy between metals, and “oxide formation energy, coordination number, alloy formation energy and ionization energy.”(1) The implemented algorithm then combed through these physical factors and analyzing which metals would be most efficient, given that certain levels of binding energy associated with metals would ascertain how fast the reaction proceeds.
In the statistical analysis, not only did they find a number of different catalytic combinations worth further investigation, but found other factors not initially prioritized also of unusual importance. For example, though they chose to focus on the physical characteristics associated with oxygen reactions, certain metal-based factors (e.g. metals reacting with other metals) actually played a role in efficiency as well to the surprise of the investigators.
Future steps for the researchers will involve a great deal of experimentation and further research with the determined catalysts and will hopefully lead to improvement in industrial catalytic processes. For now, though, we can take cues from the interdisciplinary nature of this research and perhaps continue even integrating these advanced technologies into revolutionizing the backbones of natural science today.
References and Footnotes:
- Rice University. “Machine learning can improve catalytic design: Algorithm helps chemical engineers at Rice, Penn State find hidden correlations.” ScienceDaily. ScienceDaily, 2 July 2018. <www.sciencedaily.com/releases/2018/07/180702170805.htm>.