A deep learning multi-attention Bi-GRU framework for k_cat prediction with segmentation-based insights

Area of Science:

• Biochemistry
• Computational Biology
• Machine Learning

Background:

• Enzyme catalysis is vital for metabolism, but experimental measurement of catalytic constants (kcat) lags behind rapid sequence data generation.
• This bottleneck hinders progress in metabolic engineering and synthetic biology.

Purpose of the Study:

• To introduce KcatNeuroCortex, an interpretable deep learning framework for predicting enzyme catalytic efficiency.
• To address the challenge of rapid enzyme sequence data growth versus slow experimental kcat determination.

Main Methods:

• Developed a novel deep learning architecture combining Bi-directional Gated Recurrent Units (Bi-GRU) with a multi-attention mechanism.
• Employed a segmentation-based strategy to capture local functional motifs and integrated them into a global representation.
• Focused on modeling long-range interactions influencing enzyme catalysis.

Main Results:

• KcatNeuroCortex achieved R² of 0.74 and RMSE of 0.77, a 57% improvement over DLKcat.
• Demonstrated competitive performance, especially on diverse and low-similarity enzyme sequences.
• Showcased the framework's robustness, scalability, and interpretability.

Conclusions:

• KcatNeuroCortex provides accurate and interpretable enzyme catalytic efficiency predictions.
• The framework enhances enzyme engineering and kinetic parameter estimation.
• Deep learning can offer biological insights into enzyme catalysis beyond mere prediction.