A Way Forward for Predictors of Protein Thermodynamics Stability Changes Due to Mutations in the Unavailability of Experimental Protein Structures

The technological advancement in computing facilities has necessitated its inclusion in all spheres of science. The experimental methods for estimating the impacts of mutation on protein thermodynamic stability are time-consuming and labor-intensive. However, the recent advancements in computational techniques for predicting protein thermodynamic stability caused by mutation are significant and can greatly impact the field of drug resistance and protein design and engineering. The machine learning-based methods predicting protein thermodynamic stability changes upon mutation have made a lot of advancements in the last two decades. This work highlights these recent breakthroughs for the predictors of protein thermodynamic stability without experimental protein structure. It focuses on the availability of data through various repositories, feature extraction methods for representing protein sequences, and machine learning algorithms used for the modeling. The work discusses building feature vectors of nature sequentially, evolutionarily, mutationally, and predicted structurally. The modeling details about supervised machine learning algorithms for the classification and regression task and parameter tuning information are provided. The information for the publicly available web server for the classification and regression task of protein thermo- dynamics stability changes due to mutations in the unavailability of experimental protein structures is included. The way forward for building a better predictor for protein thermodynamic stability is also suggested.