Predicting protein separation behaviour

Applying math and computers to the drug-discovery process, researchers at Rensselaer Polytechnic Institute have developed a method to predict protein separation behaviour directly from protein structure. This multi-scale protein modelling approach may reduce the time it takes to bring pharmaceuticals to market and may have significant implications for an array of biotechnology applications, including bioprocessing, drug discovery and proteomics, the study of protein structure and function.

"Predictive modelling is a new approach to drug discovery that takes information from laboratory analysis and concentrates it in predictive models that may be evaluated on a computer," said Curt M Breneman, professor of chemistry and chemical biology at Rensselaer.

The modelling technique is based on methods previously developed by Breneman's group for rapidly predicting the efficacy and side effects of small drug-like molecules. The newly developed model successfully predicted the amount of a protein that binds to a material under a range of conditions by using molecular information obtained from the protein structure. These predicted adsorption isotherm parameters then replicated experimental results by predicting the actual separation profile of proteins in chromatographic columns. Chromatography techniques are used to identify and purify molecules, in this case, particular proteins.

The researchers' computational model uses a combination of molecular-level quantitative structure-property relationship models with macroscopic steric mass action isotherm models and support vector machine regression computations.