Accelerating Metabolic Modeling: DLKcat is Now Available on Vecura

What is DLKcat?

DLKcat is a deep-learning model designed to predict the catalytic turnover number (kcat) for any given enzyme-substrate pair. By taking an enzyme's amino-acid sequence and a substrate's SMILES string as input, the model utilizes a sophisticated architecture that combines a graph neural network (GNN) for the substrate with an attention-weighted convolutional neural network (CNN) for the protein sequence. It outputs the predicted kcat value in 1/s, which represents the number of substrate molecules converted by one enzyme molecule per second under saturating conditions.

It helps researchers quickly estimate kinetic constants, which are fundamental parameters in metabolic modeling and protein engineering. It is especially useful for scientists working on genome-scale metabolic models (ecGEMs) where experimental kcat data is often sparse or unavailable.

What can users do with DLKcat on Vecura?

With DLKcat on Vecura, users can:

• Predict the turnover number (kcat) for novel enzyme-substrate combinations without needing specialized local hardware.
• Easily perform large-scale kinetic parameter estimation for entire sets of metabolic enzymes.
• Integrate automated predictions into workflows by supplying either SMILES strings or common substrate names, with built-in PubChem resolution.
• Obtain both the raw log2(kcat) regression output and the converted kcat value (1/s) for diverse downstream biological applications.

What the output means

The output provides a quantitative prediction of the enzyme's catalytic efficiency (kcat). Specifically, it delivers the kcat_value (in 1/s), the log2_kcat (useful for linear-scale statistical comparisons), and the resolved_smiles (confirming the chemical structure used for the prediction).

This output should be used to support scientific decision making. It does not replace experimental validation.

Why this matters

The turnover number (kcat) is a critical parameter for understanding enzymatic reactions and modeling metabolic pathways. However, experimentally measuring kcat is time-consuming, expensive, and limited in throughput, leading to significant gaps in our knowledge of kinetic constants across different organisms.

DLKcat bridges this gap by providing an accurate computational framework for estimating these values at a genome-scale. By offering reliable predictions, it empowers researchers to construct more robust enzyme-constrained metabolic models and accelerates the discovery of enzymes with desired catalytic properties.

• Developed by: Research group led by Li et al. (Chalmers University of Technology)
• Source: Official GitHub repository
• Reference: Li, F., et al. "Deep learning-based kcat prediction enables genome-scale enzyme-constrained metabolic modeling." Nature Catalysis (2022).