CatPred Kinetic Prediction Now Available on Vecura

What is CatPred?

CatPred is an advanced, multimodal deep-learning framework designed to predict three essential in vitro enzyme kinetic parameters—turnover number ($k_$), Michaelis constant ($K_m$), and inhibition constant ($K_i$)—directly from an enzyme's amino-acid sequence and a small-molecule substrate or inhibitor SMILES string. By leveraging representations from pretrained protein language models alongside graph-based molecular features, CatPred overcomes the traditional bottlenecks associated with experimental kinetic assays. It is particularly useful for researchers in enzyme engineering and metabolic pathway design, offering rapid, uncertainty-aware predictions even for enzyme-ligand pairs lacking prior experimental data.

What can users do with CatPred on Vecura?

With CatPred on Vecura, users can:

• Predict Kinetic Parameters: Rapidly generate $k_$ and $K_m$ values for enzymes by setting the ligand role to "substrate."
• Evaluate Inhibition Affinity: Estimate $K_i$ values for enzyme-inhibitor pairs to assess binding strength.
• Perform Batch Analysis: Process multiple enzyme-ligand pairs in a single API call for high-throughput screening.
• Assess Prediction Reliability: Utilize ensemble-based uncertainty metrics—including aleatoric and epistemic standard deviations—to gauge confidence and flag queries that are out-of-distribution.

What the output means

The output provides quantitative predictions for kinetic parameters, including both linear values (s⁻¹ or mM) for direct use in kinetic models and log₁₀-transformed values for regression analysis. Crucially, the system returns total predictive standard deviation along with specific components for data-noise (aleatoric) and model-uncertainty (epistemic). This output should be used to support scientific decision-making, such as prioritizing enzyme variants for further study. It does not replace experimental validation.

Why this matters

In enzyme engineering and synthetic biology, the "kinetic bottleneck"—where the lack of experimental data hinders the design of efficient metabolic pathways—remains a major hurdle. Traditional wet-lab assays are time-consuming and expensive, often limiting the scope of rational design.

CatPred transforms this workflow by providing an accessible, high-performance computational tool to bridge the data gap. By predicting kinetic constants in silico, researchers can filter candidate enzymes and pathways far more efficiently, accelerating the development of biocatalysts and optimized production strains for biotechnology.

• Developed by: The Maranas Research Group
• Source: Nature Communications (2025) | GitHub Repository
• Reference: Boorla, V.S., Maranas, C.D. CatPred: Deep-learning framework for predicting in vitro enzyme kinetic parameters. Nat Commun 16, [insert article number] (2025).