Pseudodata-guided Invariant Representation Learning Boosts the Out-of-Distribution Generalization in Enzymatic Kinetic Parameter Prediction

TL;DR

The paper tackles the challenge of out-of-distribution generalization in enzyme-kinetic parameter prediction by introducing O$^2$DENet, a plug-and-play module that combines biologically informed perturbations with invariant representation learning. By augmenting both enzymes (via residue masking) and substrates (via SMILES enumeration and graph masking) and enforcing consistency between raw and augmented embeddings, the approach yields robust improvements across multiple ESI predictors for $k_{cat}$ and $K_m$, especially under stringent sequence-identity-based OOD splits. Zero-shot evaluations on TAL and MS datasets demonstrate practical utility for enzyme engineering, with high success rates in identifying improved variants despite low sequence similarity to training data. Overall, O$^2$DENet provides a scalable, architecture-agnostic strategy to enhance reliability and deployability of data-driven enzyme kinetics models in real-world engineering tasks, strengthening the bridge between computational predictions and experimental design.

Abstract

Accurate prediction of enzyme kinetic parameters is essential for understanding catalytic mechanisms and guiding enzyme engineering.However, existing deep learning-based enzyme-substrate interaction (ESI) predictors often exhibit performance degradation on sequence-divergent, out-of-distribution (OOD) cases, limiting robustness under biologically relevant perturbations.We propose O$^2$DENet, a lightweight, plug-and-play module that enhances OOD generalization via biologically and chemically informed perturbation augmentation and invariant representation learning.O$^2$DENet introduces enzyme-substrate perturbations and enforces consistency between original and augmented enzyme-substrate-pair representations to encourage invariance to distributional shifts.When integrated with representative ESI models, O$^2$DENet consistently improves predictive performance for both $k_{cat}$ and $K_m$ across stringent sequence-identity-based OOD benchmarks, achieving state-of-the-art results among the evaluated methods in terms of accuracy and robustness metrics.Overall, O$^2$DENet provides a general and effective strategy to enhance the stability and deployability of data-driven enzyme kinetics predictors for real-world enzyme engineering applications.

Figures (6)

• Figure 1: The motivation and methodology of $\text{O}^2$DENet. a, The methodology of $\text{O}^2$DENet for improving the generalization ability of enzyme-substrate interaction predictors. b, The augmentation approaches of enzyme and substrate inputs, namely SMILES enumeration, molecular graph masking, and enzyme sequence mask. c, The implementation of $\text{O}^2$DENet, where augmented ESI embedding undergoes invariant learning with the raw ESI embedding as an auxiliary training task.
• Figure 2: The $R^2$-based GOOD curves of the baseline models (solid lines) and their integration with $\text{O}^2$DENet (dashed lines). Each curve represents the trend of generalization performance under different OOD levels.
• Figure 3: Ablation experiments for masking ratios on top of baseline model CatPred, where $R^2$ is adopted as the evaluation metric. ab, The ablation experiments of substrate molecule graph masking ratios for $k_{cat}$ (a) and $K_{m}$ (b) prediction tasks. cd, The ablation experiments of enzyme sequence masking ratios for $k_{cat}$ (c) and $K_{m}$ (d) prediction tasks.
• Figure S1: Ablation experiments for masking ratios on top of baseline model DLKCat, where $R^2$ is adopted as the evaluation metric. ab, The ablation experiments of substrate molecule graph masking ratios for $k_{cat}$ (a) and $K_{m}$ (b) prediction tasks. cd, The ablation experiments of enzyme sequence masking ratios for $k_{cat}$ (c) and $K_{m}$ (d) prediction tasks.
• Figure S2: Ablation experiments of enzyme sequence masking ratios on top of the baseline model UniKP, where $R^2$ is adopted as the evaluation metric. a, The ablation experiments of enzyme sequence masking ratios for $k_{cat}$. b, The ablation experiments of enzyme sequence masking ratios for $K_{m}$.