The development of robust polishing chromatographic processes is a critical step in downstream bioprocess development that can be time-consuming and resource intensive. Recently, there has been an increase in diverse protein constructs that are not amenable to platform approaches, increasing the need for novel processes to be developed for effective purification. High throughput screening (HTS) is an important tool to parse chromatographic design space and identify promising conditions to continue development. Despite its utility, HTS capabilities are challenged by tight development timelines, material scarcity, and an increasingly complex pipeline of biotherapeutics. Predictive modeling can augment HTS by leveraging historical screening data to rapidly explore and prioritize process design space, effectively expanding the range of conditions considered without the need for additional experimental screening. Here we present the development of a quantitative structure activity relationship (QSAR) model, trained from internal HTS data, that predicts protein partitioning as a function of resin and mobile phase conditions. The training dataset contains a diverse collection of screening data and has more than 8000 datapoints, covering 29 therapeutic proteins and 44 resins. The model encodes partitioning by building descriptors of the mobile phase, parameters that describe the resin, and biophysical properties of the protein. Overall, the regression model has an R