High-quality Thermodynamic Data on the Stability Changes of Proteins Upon Single-site Mutations

Abstract

We have set up and manually curated a dataset containing experimental information on the impact of amino acid substitutions in a protein on its thermal stability. It consists of a repository of experimentally measured melting temperatures (T m) and their changes upon point mutations (∆T m) for proteins having a well-resolved X-ray structure. This high-quality dataset is designed for being used for the training or benchmarking of in silico thermal stability prediction methods. It also reports other experimentally measured thermodynamic quantities when available, i.e. the folding enthalpy (∆H) and heat capacity (∆C P) of the wild type proteins and their changes upon mutations (∆∆H and ∆∆C P), as well as the change in folding free energy (∆∆G) at a reference temperature. These data are analyzed in view of improving our insights into the correlation between thermal and thermodynamic stabilities, the asymmetry between the number of stabilizing and destabilizing mutations, and the difference in stabilization potential of thermostable versus mesostable proteins.

Quantification of biases in predictions of protein stability changes upon mutations

“…They are often defined either at room temperature T r ¼ 25 C or at the melting temperature T m of the wild-type protein. Sometimes, they are not directly measured but derived from DT m measures in differential scanning calorimetry (DSC) experiments, by utilizing the fact that these two quantities are correlated, even though this is only true in a first approximation [see Pucci et al (2016) and Watson and Raleigh (2017) for further details]. All these dependencies and approximations make the datasets of the experimentally annotated mutations quite noisy, which in turn impacts the accuracy of the predictors that are trained on them.…”

Improved insights into protein thermal stability: from the molecular to the structurome scale

“…Mean values ( T m ) and root mean square deviations (σ ) of T m (in • C) for the mutations inserted in proteins with different thermal characteristics. The experimental values are derived from the original literature and are reported in a dataset [28]. The computed values are obtained using T m -HoTMuSiC [27] …”

“…Indeed, the results shown in table 1 have to be carefully checked, because they could contain biases due to the non-random sampling of the experimentally characterized mutations or to different experimental conditions at which the measurements were performed [28].…”

Role of simple descriptors and applicability domain in predicting change in protein thermostability

“…( Fig 2A ) Mutations are predominantly from apolar residues to apolar (601), polar (168) or charged (66) residues. ( Fig 2B ) dTm is confirmed [ 35 ] to significantly decrease as the fraction buried of the wild-type residue increases, where the median values were -0.5, -1.3, -4°C for mutants with ≤ 71, (71, 91), and > 91% fraction buried, respectively. ( Fig 2C ) This is to say, mutations at the surface of proteins have less effect.…”

“…The benchmark dataset used for model generation is that from Pucci et al [ 35 ] and includes 1626 high-quality curated single-point mutants from soluble proteins that have experimental dTm values in physiologically relevant conditions and reported PDB structures. 289 of the mutants are stabilizing (dTm > 1°C), 917 are destabilizing (dTm< -1°C) and 420 are neutral (-1 ≤ dTm ≤ 1°C).…”