Use DLKcat Online
Commercially Available DLKcat No-Code Web Server
DLKcat: Deep Learning Prediction of Enzyme Turnover Numbers (kcat)
DLKcat is a deep learning method that predicts enzyme turnover numbers (kcat) from a substrate structure and a protein sequence. It was reported by Feiran Li, Le Yuan, Hongzhong Lu, Gang Li, Yu Chen, Martin Engqvist, Eduard Kerkhoven and Jens Nielsen, among others, in Nature Catalysis (2022). The authors' GitHub repository provides a Matlab and Python toolbox that includes the deep learning predictor.
Experimentally measured kcat values are sparse and uneven across organisms, which limits the quality of enzyme-constrained genome-scale metabolic models (ecGEMs). Earlier pipelines relied on matching enzymes by EC number. DLKcat predicts kcat directly from sequence and substrate, so it can be applied to metabolic enzymes from any organism.
Property | Detail |
|---|---|
Inputs | Substrate (SMILES or name) and protein amino acid sequence |
Output | Predicted kcat value |
Architecture | Graph neural network for substrates, CNN for protein n-grams, attention fusion |
Training data | BRENDA and SABIO-RK: 16,838 entries, 7,822 sequences, 2,672 substrates, 851 organisms |
Venue | Nature Catalysis 5, 662-672 (2022) |
How DLKcat Predicts kcat from Substrate Structure and Protein Sequence
Substrate encoding: SMILES strings are converted to molecular graphs and processed by a graph neural network.
Protein encoding: amino acid sequences are split into overlapping n-grams and processed by a convolutional neural network.
Fusion: the two representations are concatenated and combined with an attention mechanism.
Attention for residues: attention weights can point to residues that matter for catalytic activity, useful for protein engineering.
DLKcat Performance and Applications in Enzyme-Constrained Metabolic Models
Evaluation | Result |
|---|---|
Test-set RMSE | 1.06 |
Pearson r, whole dataset | 0.88 |
Pearson r, test set | 0.71 |
Pearson r, unseen substrate or enzyme | 0.70 |
Mutated enzymes, test set | 0.78 |
The authors predicted kcat values for 343 yeast and fungal species, about 3 million enzyme-substrate pairs. Combined with a Bayesian pipeline that refines predictions using growth data, the resulting ecGEMs reached about 80% enzyme coverage and 90% reaction coverage, compared with roughly 40% and 60% for the original models, and proteome prediction error fell by about 30%.
What is Tamarind Bio?
Tamarind Bio is a no-code bioinformatics platform built to give life scientists and researchers access to powerful computational tools. Many cutting-edge machine learning models are hard to deploy and use. Tamarind provides an intuitive, web-based environment that removes the complexity of high-performance computing, software dependencies and command-line interfaces.
The platform is designed for biologists, chemists and other researchers who may not have a background in programming or cloud infrastructure but want to run models on their own data. Key features include:
A user-friendly graphical interface for setting up and launching experiments
A robust API for integration into existing research pipelines
An automated system for managing and scaling computational resources
Tamarind treats information and data security as a top priority, as detailed in its Trust Center and Terms of Service.
Accelerating Discovery with DLKcat on Tamarind Bio
Metabolic modeling: parameterize enzyme-constrained genome-scale models without relying on EC-number matching.
Protein engineering: compare predicted kcat across mutants and identify high-attention residues that affect activity.
Promiscuity and underground metabolism: rank preferred, alternative and underground substrates for an enzyme.
Enzyme evolution and phenotype studies: compare kcat across species, as the authors did for the Crabtree effect in yeasts.
How to Use DLKcat on Tamarind Bio
Open the tool: log in to tamarind.bio and open DLKcat.
Provide the enzyme: enter the protein amino acid sequence.
Provide the substrate: enter a SMILES string, or a substrate name if exposed.
Batch inputs: supply several enzyme-substrate pairs at once, if exposed.
Run the job: submit the prediction.
Download outputs: retrieve the predicted kcat values.
Interpret: treat values as relative estimates and compare variants or substrates side by side.
Use downstream: feed predictions into ecGEM construction or enzyme ranking.
Things to Keep in Mind
Training data are mostly in vitro, so predictions may differ from in vivo kcat values.
Predicted activity toward random substrates is likely too high because negative data are scarce.
pH, temperature and other environmental factors are not modeled.
Multi-substrate reactions and heteromeric complexes are simplified by taking the maximum predicted kcat.
Source: Li, F., Yuan, L., Lu, H., Li, G., Chen, Y., Engqvist, M.K.M., Kerkhoven, E.J. and Nielsen, J. (and others), Deep learning-based kcat prediction enables improved enzyme-constrained model reconstruction. Nature Catalysis 5, 662-672 (2022). DOI: 10.1038/s41929-022-00798-z. Code: https://github.com/SysBioChalmers/DLKcat