A group at Oak Ridge national laboratory is developing an automated system that studies the performance of elemental chemical separations via extraction chromatography. Performing extraction chromatography by hand is a time-consuming process and it is of interest to speed up this process. We describe an ongoing literature survey that data mined nearly 10,000 individual separations. The database indexes 41 different resins, 2 acids, 77 elements (focusing mainly on transition metals, actinides, and lanthanides), and various acid concentrations which vary by 5 orders of magnitude. In addition to those independent variables, the data base includes a dependent variable which is the coefficient of separation efficiency or performance referred to as the distribution coefficient and stylized as K_d. This distribution coefficient is useful in designing a series of separations that maximally separate different elemental and molecular species for enhanced downstream analysis, medical treatments, use as a purified feedstock for manufacturing, as well as for many other uses. This work focuses on the use of machine learning to estimate K_d value from the independent variables. The poster will describe the results as well as discuss some of the challenges of the datasets, such as sparsity.
