Stability and thermostability

Thermostability refers to a protein’s ability to remain folded at high temperatures or under harsh conditions. It is a highly desirable property for engineered proteins.

Prediction

• Phantom epistasis refers to the inclusion of unnecessary model parameters when building biophysical/statistical fitness models (Fitness prediction). Faure et al. (1) attribute this to the epistasis mechanisms reviewed by Domingo et al. (2).
• Thermodynamic reversability can be used for expanding training sets for stability prediction/ddG prediction ML models. However, it has been shown to lead to biases that favor WT amino acids. Diaz et al. (3) claim to mitigate this.
• The amount of ddG data available for a given residue for training can be expanded using thermodynamic permutation, where $n$ measurements are increased to $n\ast (n-1)$. This was used by MutComputeXGT on the Tsuboyama et al. (4) dataset. It is useful for stability prediction and improves generalization in (3).
• ddG data is skewed with hydrophobic amino acids (e.g., alanine scans). This has been reported to increase solvation ddG by 0.8 kcal/mol in studies cited by (3). The Tsuboyama et al. (4) data does not have this bias.