Learning deep representations of enzyme thermal adaptation

Abstract

Temperature is a fundamental environmental factor that shapes the evolution of organisms. Learning thermal determinants of protein sequences in evolution thus has profound significance for basic biology, drug discovery, and protein engineering. Here, we use a dataset of over 3 million enzymes labeled with optimal growth temperatures (OGT) of their source organisms to train a deep neural network model (DeepET). The protein-temperature representations learned by DeepET provide a temperature-related statistical summary of protein sequences and capture structural properties that affect thermal stability. For prediction of enzyme optimal catalytic temperatures and protein melting temperatures via a transfer learning approach, our DeepET model outperforms classical regression models trained on rationally designed features and other recent deep-learning-based representations. DeepET thus holds promise for understanding enzyme thermal adaptation and guiding the engineering of thermostable enzymes.

High-Temperature Tolerance Protein Engineering through Deep Evolution

“…These measurements showed that 78.0% (recall) of all labeled HTTPs were predicted as HTTPs by the model, and 86.0% (precision) of all sequences predicted as HTTPs by the model were actually the labeled HTTPs in the test set (Table 1 ). Unlike state-of-the-art methods in protein thermostability prediction that utilize OGT data, such as DeepET [ 25 ] or Tome [ 26 ], our model does not directly predict metrics related to thermostability. Instead, it classifies sequences exhibiting HTTP traits, subsequently leveraging this classification as a pression in a virtual evolutionary process.…”

“…In contrast with conventional stability prediction methods relying on scarce high-resolution data, our technique leverages large amount of OGT data from genomic sequences across diverse organisms. Some prior efforts have utilized OGT data to achieve moderate protein thermostability discrimination [ 25 , 27 ]. For instance, TemStaPro [ 27 ] exhibits higher accuracy and recall than our Thermo-selector.…”

“…From a data perspective, procuring data annotated with organisms’ optimal growth temperature (OGT) labels is relatively more straightforward compared to obtaining highly precise protein thermostability property labels. Several OGT-based methods have been developed to identify thermostable proteins [ 25 – 27 ]. However, the relatively ambiguous nature of these OGT data labels, which do not demonstrate a one-to-one correspondence with protein properties, limits the accuracy of these methods.…”