Nonsynonymous single-nucleotide polymorphisms often result in altered protein stability while playing crucial roles both in the evolution process and in the development of human diseases. Prediction of change in the thermodynamic stability due to such missense mutations will help in protein engineering endeavors and will contribute to a better understanding of different disease conditions. Here, we develop a machine-learning-based framework, viz., ProTSPoM, to estimate the change in protein thermodynamic stability arising out of single-point mutations (SPMs). ProTSPoM outperforms existing methods on the S2648 and S1925 databases and reports a Pearson correlation coefficient of 0.82 (0.88) and a root-mean-squared-error of 0.92 (1.06) kcal/mol between the predicted and experimental ΔΔG values on the long-established S350 (tumor suppressor p53 protein) data set. Further, we estimate the change in thermodynamic stability for all possible SPMs in the DNA binding domain of the p53 protein. We identify single-nucleotide polymorphisms in p53 which are plausibly detrimental to its structural integrity and interaction affinity with the DNA molecule. ProTSPoM with its reliable estimates and time-efficient prediction is well suited to be integrated with existing protein engineering techniques. The ProTSPoM web server is accessible at http://cosmos.iitkgp.ac.in/ProTSPoM/.