Identification of Thermophilic Proteins Based on Sequence-Based Bidirectional Representations from Transformer-Embedding Features

Thermophilic proteins have great potential to be utilized as biocatalysts in biotechnology. Machine learning algorithms are gaining increasing use identifying such enzymes, reducing or even eliminating the need for experimental studies. While most previously used machine methods were based on manually designed features, we developed BertThermo, a model using Bidirectional Encoder Representations from Transformers (BERT), an automatic feature extraction tool. This method combines variety of and engineering methods, while relying single-feature encoding protein sequence alone input. BertThermo achieved accuracy 96.97% 97.51% 5-fold cross-validation independent testing, respectively, thermophilic more reliably than any described predictive algorithm. Additionally, was tested by balanced dataset, imbalanced dataset with homology sequences, results show that best robustness comparied state-of-the-art methods. The source code is available.