N. Sukumar1, Curt M Breneman1, Steven M. Cramer2, James A. Moore3, Kristin P. Bennett4, Mark J. Embrechts5, Min Li1, Jia Liu2, and Long Han6. (1) Department of Chemistry and Center for Biotechnology and Interdisciplinary Studies, Rensselaer Polytechnic Institute, 110 8th St., Troy, NY 12180-3590, (2) Department of Chemical and Biological Engineering, Rensselaer Polytechnic Institute, Troy, NY 12180, (3) Department of chemistry, Rennselaer Polytechnic Institute, 110-8th street, Troy, NY 12180, (4) Department of Mathematics, Rensselaer Polytechnic Institute, Amos Eaton Building, 110 8th St, Troy, NY 12180, (5) Decision Sciences and Engineering Systems, Rensselaer Polytechnic Institute, 110 8th St, Troy, NY 12180, (6) Decision Science and Engineering Systems, RPI, 110 8th St, Troy, NY 12180
Low-molecular-weight displacers employed in ion-exchange displacement chromatography have shown a great potential for the purification of proteins from complex mixtures. One of the advantages being their ability to carry out selective displacement chromatography in which target proteins can be eluted separately. Identifying efficient displacers, however, is a major challenge for protein displacement chromatography, as it depends not only on the protein mixtures, but also on the chemistry of the stationary phase and the conditions of the mobile phase. The choice of displacers is still mostly driven by trial-and-error and is largely dependent on domain knowledge of an expert. In this work we investigate an efficient procedure to quickly predict novel selective displacers: a small set of known selective displacers are used to train machine learning models (SVM and decision trees) that are then used to identify novel selective displacers from available commercial chemical catalogs and to progressively enrich the models.