Predicting a protein's melting temperature from its amino acid sequence.

Melting temperature is an important characteristic feature of a protein and is used for various purposes such as in drug development. Currently protein melting temperature is determined by laboratory methods such as Differential Scanning Calorimetry, Circular Dichroism, Fourier transform infrared spectroscopy and several other methods. These methods are laborious and costly. Therefore, we propose a novel bioinformatics based method for predicting protein melting temperature from amino acid sequence of a protein. This is not only a challenging task but has been previously unexplored. For this study, melting temperature of 230 proteins from a range of organisms was collected along with their sequence information from the published literature. The melting temperature of these proteins represents a very large spectrum and varies between 25°C and 113°C. The protein sequences are then used to derive two sets of sequence-driven features, namely amino acid composition (AAC) and pseudo-amino acid composition (PseudoAAC) to characterise the proteins. In order to predict the melting temperature, two different computational intelligence methods, namely artificial neural networks (ANN) and adaptive network-fuzzy inference system (ANFIS) were utilized. Amongst over 100 different models generated, the ANN produced the best model with the least error (0.01087 for the AAC and 0.01086 for the pseudoAAC). As both feature sets yielded quite similar error and computation of pseudoAAC is costly when compared to that of AAC, traditional AAC seems to be an effective feature set for predicting melting temperature. The results obtained in this study are very promising and, for the first time, shows that the melting temperature of a protein can be predicted from its amino acid sequence only. Therefore, costly lab-based experiments may not be required to measure the melting temperature and the bioinformatics models can help speed up laboratory processes such as in drug development.