AdventML: Advanced Enzyme Temperature Prediction with Transformer-Based Embeddings and Resampling Strategies

Accurate prediction of enzymes’ optimal catalytic temperature ( T _opt ) is crucial in biotechnology, as enzymes with extreme T _opt values are highly desirable for reactions at extreme temperatures and for their general stability. However, experimental determination of T _opt is costly, labor-intensive, and time-consuming. Meanwhile, existing computational methods suffer from small and imbalanced datasets, subop-timal predictions at extreme temperatures, and insufficient validation.

In this study, we address these challenges by expanding the T _opt dataset and validating on an independent test set based on sequence similarity. We further tackle these limitations by comparing multiple resampling techniques to improve predictions at extremes and by considering diverse protein rep-resentations and multiple machine learning architectures. Overall, the best performing models reached R ² ≈ 0.64 with MAE ≈ 7–8 °C, while extreme resampling improved tail performance (reducing tail MAE by up to ~1.8 °C). Notably, our models show improved performance over state-of-the-art prediction models. We also demonstrate that accurate prediction of T _opt is achievable even in the absence of organ-ism growth temperature (OGT). Our T _opt prediction models are made freely available as AdventML on GitHub.