Background: Computational methods of predicting protein stability changes upon missense mutations are invaluable tools in high-throughput studies involving a large number of protein variants. However, they are limited by a wide variation in accuracy and difficulty of assessing prediction uncertainty. Using a popular computational tool, FoldX, we develop a statistical framework that quantifies the uncertainty of predicted changes in protein stability. Results: We show that multiple linear regression models can be used to quantify the uncertainty associated with FoldX prediction for individual mutations. Comparing the performance among models with varying degrees of complexity, we find that the model precision improves significantly when we utilize molecular dynamics simulation as part of the FoldX workflow. Based on the model that incorporates information from molecular dynamics, biochemical properties, as well as FoldX energy terms, we can generally expect upper bounds on the uncertainty of folding stability predictions of ± 2.9 kcal/mol and ± 3.5 kcal/mol for binding stability predictions. The uncertainty for individual mutations varies; our model estimates it using FoldX energy terms, biochemical properties of the mutated residue, as well as the variability among snapshots from molecular dynamics simulation. Conclusions: Using a linear regression framework, we construct models to predict the uncertainty associated with FoldX prediction of stability changes upon mutation. This technique is straightforward and can be extended to other computational methods as well.
Here we present a novel protocol for the construction of saturation single-site—and massive multi-site—mutant libraries of a bacteriophage. We segmented the ΦX174 genome into 14 non-toxic and non-replicative fragments compatible with golden gate assembly. We next used nicking mutagenesis with oligonucleotides prepared from unamplified oligo pools with individual segments as templates to prepare near-comprehensive single-site mutagenesis libraries of genes encoding the F capsid protein (421 amino acids scanned) and G spike protein (172 amino acids scanned). Libraries possessed greater than 99% of all 11,860 programmed mutations. Golden Gate cloning was then used to assemble the complete ΦX174 mutant genome and generate libraries of infective viruses. This protocol will enable reverse genetics experiments for studying viral evolution and, with some modifications, can be applied for engineering therapeutically relevant bacteriophages with larger genomes.
Many systems exhibit multistage mechanisms where failure of any single component leads to overall failure. Standard logistic regression, with its additive log-odds structure, is not well suited for such data-generating processes. We propose the Multistage Binomial (MSB) model, an extension of the binomial generalized linear model in which component effects combine multiplicatively on the success-probability scale. The MSB model naturally accounts for unobserved components by allowing the success probability to asymptote below one. It additionally incorporates measurement variability in explanatory variables through a Berkson-type error framework. We establish conditions for identifiability, develop a penalized maximum likelihood estimation procedure, and a non-standard likelihood ratio test for unit asymptote. Using synthetic data on mutated protein viability, we show that the likelihood ratio test is conservative, and only rejects the unit asymptote assumption when it is strongly supported by data. We also demonstrate that the MSB model provides more accurate inference and prediction than traditional logistic regression in multistage settings. Our simulations further show that larger sample sizes are required with increased proportion of unobserved components. We provide a chart for sample size determination under an MSB design for real data analysis given the desired accuracy and the distribution of predictors.
The relationship between genotype and phenotype underlies our ability to understand and predict evolution. Efforts to build genotype-phenotype (GP) maps have revealed several unifying rules: epistasis is pervasive, fitness effects are not normally distributed, and the GP map is non-linear and complicated for high-level phenotypes. A critical step in developing GP maps is evaluating how well predictive models do in explaining observed phenotypes. We utilize the simplicity of a bacteriophage (ΦX174) study system to test if an intermediate phenotype (predicted stability of the G capsid protein) explains more complex phenotypes. In doing so, we compare the predictive performance of free energies of folding and binding obtained using numerous molecular modeling methods as well as phylogenetic and basic biochemical/biophysical properties of amino acid substitutions. By creating a large mutational library, we find that ΦX174 tolerates about 50% of the amino acid substitutions we inserted into the G protein and that molecular modeling compliments other substitution models for predicting viability. Mutations predicted to have large destabilizing effects are especially informative and are almost universally detrimental. These large-effect substitutions often coincide with the most conserved residues in the G protein. Apart from large-effect mutations, our ability to predict ΦX174 phenotypes is fairly poor and we explore various potential confounding factors (e.g., codon bias) that could be considered to improve viability predictions.
Background: Protein function depends on the stable folding or binding of peptide chains, and the extent to which an amino acid substitution disrupts this stability can be a strong predictor of the mutation’s impact on the protein’s function and the organism’s fitness. This study seeks to understand this relationship in bacteriophage ϕX174 using an experimental technique of deep mutational scanning and computationally predicted protein stability. Results: Analyzing the viability data using a newly developed multistage binomial model confirms that highly destabilizing mutations predict inviability. For single-site variants, the model predicts that ΔΔG of folding greater than ~3 kcal/mol and ΔΔG of binding greater than ~6 kcal/mol cuts the probability of survival by half of the maximum. The maximum probability, even for minimal ΔΔG values, plateaus at around 87%, reflecting the influence of unobserved factors. Fitting of double-site variants data shows similar results. In contrast, a linear regression analysis of viable mutants reveals that the effect of protein stability is too weak to be a reliable predictor of fitness. Conclusion: In this study, we introduce a novel modeling technique to explore the relationship between protein stability and survival outcomes. While large destabilization often indicates inviability, most destabilizing mutations exert only small to moderate effects and do not reliably predict survival. Although protein stability is a crucial biophysical property influencing biological systems, it represents just one of multiple factors that contribute to fitness and survival.