Use CatPred Online
Commercially Available CatPred No-Code Web Server
CatPred: Deep Learning Prediction of Enzyme Kinetic Parameters (kcat, Km and Ki) with Uncertainty
CatPred is a machine learning framework for predicting in vitro enzyme kinetic parameters, namely turnover number (kcat), Michaelis constant (Km) and inhibition constant (Ki), with uncertainty estimates. According to the project repository it was developed by Veda Sheersh Boorla and Costas D. Maranas and published in Nature Communications in 2025. The repository is released under an MIT license.
Enzyme kinetic parameters are essential for metabolic engineering, enzyme design and systems biology, yet experimental measurements are available for only a small fraction of known enzymes. Prior predictors are often limited to kcat or Km and may lose accuracy on enzymes unlike those in the training data.
CatPred addresses this with probabilistic regression that outputs both a prediction and a variance, along with a larger curated dataset, so that users can judge how much to trust each result.
Property | Detail from the paper |
|---|---|
Inputs | Enzyme amino acid sequence (or 3D structure) and substrate SMILES |
Outputs | kcat, Km or Ki prediction with aleatoric and epistemic uncertainty |
Enzyme features | Sequence attention, ESM-2 language model features (1280-dimensional), E-GNN structure embeddings (128-dimensional) |
Substrate features | Directed message passing neural network on RDKit atom-bond graphs |
Dataset (CatPred-DB) | 23,197 kcat, 41,174 Km and 11,929 Ki entries from BRENDA and SABIO-RK |
Runtime | About 20 seconds on CPU and 10 seconds on GPU |
How CatPred Predicts Enzyme Kinetics and Uncertainty
Encode the enzyme: sequence attention, ESM-2 features and optional structure embeddings describe the protein.
Encode the substrate: a directed message passing network processes the 2D molecular graph. For kcat, SMILES of all reactants are concatenated; for Km and Ki only the relevant substrate is used.
Probabilistic regression: the network outputs a mean and variance of the log10-transformed parameter, trained with a negative log-likelihood loss.
Ensemble uncertainty: an ensemble of 10 models gives the prediction as the average of means, with aleatoric uncertainty from the mean variance and epistemic uncertainty from the spread of ensemble means.
CatPred Accuracy on Held-Out and Out-of-Distribution Enzymes
Parameter | Held-out test R2 | Out-of-distribution R2 | Within one order of magnitude (held-out / OOD) |
|---|---|---|---|
kcat | 0.607 | 0.390 | about 80% / 63.5% |
Km | 0.648 | 0.536 | about 87% / 82.7% |
Ki | 0.637 | 0.409 | about 70% / 58.6% |
Confidence-based filtering
Restricting to the most confident quarter of predictions raises R2 to 0.78 for kcat, 0.76 for Km and 0.61 for Ki, with mean absolute error dropping by about 36% relative to the full set. As an example, the predicted Km for human hexokinase with D-glucose was 5.58 mM, within 7% of the reported 6.3 mM.
Comparison with other tools
On out-of-distribution kcat, CatPred reached R2 of 0.39 versus 0.40 for TurNup, while UniKP scored 0.65 versus 0.61 for CatPred in-distribution. The authors cite a negative R2 of -0.61 for DLKcat on dissimilar sequences, as reported by Kroll et al.
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 CatPred on Tamarind Bio
Enzyme screening: rank candidate enzymes by predicted kcat and Km before wet-lab assays.
Metabolic modeling: fill gaps in kinetic parameters for pathway and systems-biology models.
Inhibitor studies: estimate Ki values for enzyme-inhibitor pairs.
Trust-aware prioritization: use uncertainty to focus on high-confidence predictions.
How to Use CatPred on Tamarind Bio
Log in and open the tool: sign in to tamarind.bio and select CatPred.
Enter the enzyme: provide the amino acid sequence, or a structure if the tool accepts one.
Enter the substrate: provide the substrate SMILES; for Ki, the inhibitor.
Select the parameter: choose kcat, Km or Ki, if exposed.
Run the prediction: submit the job.
Download results: retrieve predictions and uncertainty values.
Interpret: predictions are log10-scale; favor those with low uncertainty.
Use downstream: feed values into models or enzyme selection decisions.
Things to Keep in Mind
Quality appears limited mainly by experimental noise in the underlying data.
Out-of-distribution performance for kcat and Ki is weaker than in-distribution.
The Ki dataset is small, and language model features appear to overfit it on out-of-distribution sets.
Conditions such as pH and temperature are not modeled, and engineered enzymes were excluded.
Source: Boorla V. S., Maranas C. D. "CatPred: a comprehensive framework for deep learning in vitro enzyme kinetic parameters." Nature Communications 16, 2072 (2025), DOI 10.1038/s41467-025-57215-9; preprint https://www.biorxiv.org/content/10.1101/2024.03.10.584340v2. Code: https://github.com/maranasgroup/CatPred.