Penn has been a leader in the development of High Throughput Experimentation in academia. Initial efforts were in collaboration with Merck leading to Penn’s own High Throughput Experimentation Center located in the Chemistry Department. Reactions are conducted in parallel using microtiter plates in either 4-well or 96-well formats with reaction volumes of 10, 25, of 100 µL (1-10 µM) and outcomes (chemical yield vs an internal standard, selectivity among several products, etc.) are analyzed using a range of equipment through automated sampling (GC, GC-MS, HPLC, UPLC, UPLC-MS, and SFC). While some robotic equipment is present to enable solid handling and kinetics analyses, routine screens are prepared using multi-channel pipetters. This low-barrier approach means that any researcher in the department can quickly learn and execute a screen with minimal training (no large robotic manuals needed!). As a result, many users take advantage of the facility and gain expertise, which translates into enhanced employment opportunities down the line as many companies are setting up or expanding such facilities.

In our research, we use HTE to screen reaction conditions. This technique can be used in routine reaction optimization (e.g. find the right ligand, base, metal source for a Suzuki reaction) or in new reaction development. For the latter, this effort can sometimes be straightforward, but till requires wise selection of diversity in the different screening elements. Indeed, it is all too easy to run an entire screen and get 0% yield across the board which is not useful. One example of our work shows how frighteningly narrow the successful parameters can be. The technique can also be used in concert with a broader hypothesis to discover fundamental new reactions. In the context of oxidative coupling chemistry, we discovered a chromium catalyst that is particularly effective in phenol-phenol cross coupling.

Most recently, we have been using the HTE center to collect a large number of data points for reactions in order to engage in data mining vs supervised and unsupervised learning. In the context of unsupervised learning, relationships between the data are generated by clustering. Further work with both types of machine learning are underway.