Development of protein moment descriptors and pH-dependent descriptors for prediction of protein affinity in hydrophobic interaction chromatography systems


Qiong Luo1, Asif Ladiwala2, Dechuan Zhuang1, N Sukumar1, Curt M Breneman1, and Steve M. Cramer2. (1) Department of Chemistry, Rensselaer Polytechnic Institute, Cogswell 306, 110 8th St, Troy, NY 12180, (2) Department of Chemical and Biological Engineering, Rensselaer Polytechnic Institute, 110 8th Street, Troy, NY 12180
Hydrophobic Interaction Chromatography (HIC) is commonly employed in the biotech industry for the downstream processing of proteins and other biomolecules. The selectivity of this technique can be optimized by varying the composition of the stationary phase as well as the pH of the mobile phase. In the present work, the effect of resin chemistry on binding affinity of proteins in HIC are investigated using high-throughput experimentation and Quantitative Structure-Retention Relationship (QSRR) modeling. Linear gradient experiments were carried out for 36 proteins on four different HIC resins having different backbone and ligand chemistries ĘC namely Phenyl Sepharose, Butyl Sepharose, Phenyl 650M and Butyl 650M. A number of sets of novel protein descriptors are developed in this study, including moment descriptors and pH-dependent descriptors, which are based on RECON/TAE method and MOE descriptors. In the development of protein moment descriptors, moments of various physico-chemical property distributions of proteins up to and including second order are calculated based on protein crystal structures using either all the protein atoms or only surface atoms identified by Delaunay Tessellation. Restricting the descriptors to surface atoms eliminates the contributions of atoms on deeply buried residues. Support Vector Machine (SVM) regression has been employed to obtain predictive QSRR models. The predictive ability of these models are verified for a randomly selected test set of proteins not included in the training of the model. The relative importance of each selected descriptor in the final models are provided by star plot analysis and correlation matrices. Once these predictive models have been validated, they can be used as an automated prediction tool for Virtual High-Throughput Screening (VHTS).

8:00 PM-10:00 PM, Monday, August 23, 2004 Pennsylvania Convention Center -- Hall D, Sci-Mix

Division of Chemical Information

The 228th ACS National Meeting, in Philadelphia, PA, August 22-26, 2004