Accurate predictions of protein stability have great potential to accelerate progress in computational protein design, yet the correlation of predicted and experimentally determined stabilities remains a significant challenge. To address this problem, we have developed a computational framework based on negative multistate design in which sequence energy is evaluated in the context of both native and non-native backbone ensembles. This framework was validated experimentally with the design of ten variants of streptococcal protein G domain β1 that retained the wild-type fold, and showed a very strong correlation between predicted and experimental stabilities (R2 = 0.86). When applied to four different proteins spanning a range of fold types, similarly strong correlations were also obtained. Overall, the enhanced prediction accuracies afforded by this method pave the way for new strategies to facilitate the generation of proteins with novel functions by computational protein design.
Submitter: Barry Olafson
Submission Date: July 31, 2017, 11:46 a.m.
|Number of data points||136|
|Assays/Quantities/Protocols||Experimental Assay: Cm ; Derived Quantity: SD of Cm ; Derived Quantity: % difference Cm(predicted,neg) from Cm(expt) ; Derived Quantity: % difference Cm(predicted,pos) from Cm(expt) ; Derived Quantity: % difference Cm(regression,neg) from Cm(expt) ; Derived Quantity: % difference Cm(regression,pos) from Cm(expt) ; Computational Protocol: Cm(regression, negative design) ; Computational Protocol: Cm(predicted, positive design) ; Computational Protocol: Cm(predicted, negative design) ; Computational Protocol: Cm(regression, positive design)|
|Libraries||Experimental, predicted, and regression Cms for GB1 test set (positive and negative MSD)(Table 2) ; Experimental and regression Cms for GB1 training set (positive and negative MSD) (Table 1)|