Developing a platform for in-silico protein design: Applications to thermostability QSAR modeling
Informed protein design enables better biologics and green chemistry. To balance efficiency and accuracy, prudent design methodologies incorporating sufficient and necessary information are required. Descriptors used to predict changes in protein thermostability ranged from simple sequenced based to more costly energy terms. Using three subsets of the proTherm database, classification and regression quantitative structure relationship models were developed. Classification models enhanced predictive power. Although energy terms were deemed most important in a random forest model, MOE protein descriptors alone were sufficient for thermostability predictive enhancement in an industry setting.