CoChat
Sign Up
Cited by 226Machine Learning in BioinformaticsEnzyme Catalysis and ImmobilizationProtein Structure and Dynamics

The DOI matches its Crossref record; 53 of the 54 references it cites were found in a registry or an index.

UniKP: a unified framework for the prediction of enzyme kinetic parameters

Open access
Yu, Han, Deng, Huaxiang, He, Jiahui, Keasling, Jay D., Luo, Xiaozhou·2023·Nature Communications
DOI 10.1038/s41467-023-44113-1PMC PMC10713628

Published version of a preprint

Open this paper in CoChat — free account

Abstract

Prediction of enzyme kinetic parameters is essential for designing and optimizing enzymes for various biotechnological and industrial applications, but the limited performance of current prediction tools on diverse tasks hinders their practical applications. Here, we introduce UniKP, a unified framework based on pretrained language models for the prediction of enzyme kinetic parameters, including enzyme turnover number ( k cat ), Michaelis constant ( K m ), and catalytic efficiency ( k cat / K m ), from protein sequences and substrate structures. A two-layer framework derived from UniKP (EF-UniKP) has also been proposed to allow robust k cat prediction in considering environmental factors, including pH and temperature. In addition, four representative re-weighting methods are systematically explored to successfully reduce the prediction error in high-value prediction tasks. We have demonstrated the application of UniKP and EF-UniKP in several enzyme discovery and directed evolution tasks, leading to the identification of new enzymes and enzyme mutants with higher activity. UniKP is a valuable tool for deciphering the mechanisms of enzyme kinetics and enables novel insights into enzyme engineering and their industrial applications.

Overview

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1038/s41467-023-44113-1

Cites

53 of 53 references verified0 citation errors · 0 not checkable

Statistics

Figures

Fig. 1 The overview of UniKP. a Enzyme sequence representation module: Information about enzymes was encoded using a pretrained language model, ProtT5-XL-UniRef50. Each amino acid was converted into a 1024-dimensional vector on the last hidden layer, and the resulting vectors were summed and averaged by mean pooling, generating a 1024-dimensional vector to represent the enzyme. b Substrate structure representation module: Information about substrates was encoded using a pretrained language model, SMILES Transformer model. The substrate structure was converted into a simplified molecular-input line-entry system (SMILES) representation and input into a pretrained SMILES transformer to generate a 1024-dimensional vector. This vector was generated by concatenating the mean and max pooling of the last layer, along with the first outputs of the last and penultimate layers. c Machine learning module: An explainable Extra Trees model took the concatenated representation vector of both the enzyme and substrate as input and generated a predicted kcat, Km or kcat / Km value. d EF-UniKP: A framework that considers environmental factors to generate an optimized prediction. It is validated on two representative datasets: pH and temperature datasets. e Various re-weighting methods were used to adjust the sample weight distribution to generate an optimized prediction for high-value prediction task.
Fig. 1 The overview of UniKP. a Enzyme sequence representation module: Information about enzymes was encoded using a pretrained language model, ProtT5-XL-UniRef50. Each amino acid was converted into a 1024-dimensional vector on the last hidden layer, and the resulting vectors were summed and averaged by mean pooling, generating a 1024-dimensional vector to represent the enzyme. b Substrate structure representation module: Information about substrates was encoded using a pretrained language model, SMILES Transformer model. The substrate structure was converted into a simplified molecular-input line-entry system (SMILES) representation and input into a pretrained SMILES transformer to generate a 1024-dimensional vector. This vector was generated by concatenating the mean and max pooling of the last layer, along with the first outputs of the last and penultimate layers. c Machine learning module: An explainable Extra Trees model took the concatenated representation vector of both the enzyme and substrate as input and generated a predicted kcat, Km or kcat / Km value. d EF-UniKP: A framework that considers environmental factors to generate an optimized prediction. It is validated on two representative datasets: pH and temperature datasets. e Various re-weighting methods were used to adjust the sample weight distribution to generate an optimized prediction for high-value prediction task.CC BY · PubMed Central
Fig. 2 Performance comparison of different models. Comparison of Root Mean Square Error (RMSE) (a), Pearson Correlation Coefficient (PCC) (b), Mean Absolute Error (MAE) (c), and R2 (Coefficient of Determination) (d) values between experimentally measured kcat values and predicted kcat values of 16 diverse machine learning models and 2 deep learning models. The kcat values of all samples were predicted independently using 5-fold cross-validation. Each bar in the graph represents the models’ performance with respect to this metric. The “Extra Trees” model is highlighted in yellow, while other models are depicted in blue. The corresponding numerical values for each bar are provided on the right side. Source data are provided as a Source Data file.
Fig. 2 Performance comparison of different models. Comparison of Root Mean Square Error (RMSE) (a), Pearson Correlation Coefficient (PCC) (b), Mean Absolute Error (MAE) (c), and R2 (Coefficient of Determination) (d) values between experimentally measured kcat values and predicted kcat values of 16 diverse machine learning models and 2 deep learning models. The kcat values of all samples were predicted independently using 5-fold cross-validation. Each bar in the graph represents the models’ performance with respect to this metric. The “Extra Trees” model is highlighted in yellow, while other models are depicted in blue. The corresponding numerical values for each bar are provided on the right side. Source data are provided as a Source Data file.CC BY · PubMed Central
Fig. 3 High accuracy of UniKP in enzyme kcat prediction. a Comparison of average coefficient of determination (R²) values for DLKcat and UniKP after five rounds of random test set splitting (n = 1684). b Comparison of the root mean square error (RMSE) between experimentally measured kcat values and predicted kcat values of DLKcat and UniKP for training (n = 15,154) and test sets (n = 1684). Dark bars represent RMSE of DLKcat and light bars for UniKP. c Scatter plot illustrating the Pearson coefficient correlation (PCC) between experimentally measured kcat values and predicted kcat values of UniKP for the test set (N = 1684), showing a strong linear correlation. The color gradient represents the density of data points, ranging from blue (0.02) to red (0.28). d Comparison of RMSE between experimentally measured kcat values and predicted kcat values of DLKcat and UniKP in various experimental kcat numerical intervals. Dark bars represent RMSE of DLKcat and light bars for UniKP. e Enzymes with significantly different kcat values between primary central and energy metabolism, and intermediary and secondary metabolism. An independent two-sided t-test to determine whether the means of two independent samples differ significantly. Primary central and energy metabolism (n = 3098) and intermediary and secondary metabolism (n = 4201) were examined in this analysis. f Shapley additive explanations (SHAP) analysis for the top 20-feature Extra Trees model. The impact of each feature on kcat values is illustrated through a swarm plot of their corresponding SHAP values. The color of the dot represents the relative value of the feature in the dataset (high-to-low depicted as red-to-blue). The horizontal location of the dots shows whether the effect of that feature value contributed positively or negatively in that prediction instance (x-axis). In each box plot (a, e), the central band represents the median value, the box represents the upper and lower quartiles and the whiskers extend up to 1.5 times the interquartile range beyond the box range. Source data are provided as a Source Data file.
Fig. 3 High accuracy of UniKP in enzyme kcat prediction. a Comparison of average coefficient of determination (R²) values for DLKcat and UniKP after five rounds of random test set splitting (n = 1684). b Comparison of the root mean square error (RMSE) between experimentally measured kcat values and predicted kcat values of DLKcat and UniKP for training (n = 15,154) and test sets (n = 1684). Dark bars represent RMSE of DLKcat and light bars for UniKP. c Scatter plot illustrating the Pearson coefficient correlation (PCC) between experimentally measured kcat values and predicted kcat values of UniKP for the test set (N = 1684), showing a strong linear correlation. The color gradient represents the density of data points, ranging from blue (0.02) to red (0.28). d Comparison of RMSE between experimentally measured kcat values and predicted kcat values of DLKcat and UniKP in various experimental kcat numerical intervals. Dark bars represent RMSE of DLKcat and light bars for UniKP. e Enzymes with significantly different kcat values between primary central and energy metabolism, and intermediary and secondary metabolism. An independent two-sided t-test to determine whether the means of two independent samples differ significantly. Primary central and energy metabolism (n = 3098) and intermediary and secondary metabolism (n = 4201) were examined in this analysis. f Shapley additive explanations (SHAP) analysis for the top 20-feature Extra Trees model. The impact of each feature on kcat values is illustrated through a swarm plot of their corresponding SHAP values. The color of the dot represents the relative value of the feature in the dataset (high-to-low depicted as red-to-blue). The horizontal location of the dots shows whether the effect of that feature value contributed positively or negatively in that prediction instance (x-axis). In each box plot (a, e), the central band represents the median value, the box represents the upper and lower quartiles and the whiskers extend up to 1.5 times the interquartile range beyond the box range. Source data are provided as a Source Data file.CC BY · PubMed Central
Fig. 4 UniKP markedly discriminates kcat values of enzymes and their mutants. Scatter plot illustrating the Pearson coefficient correlation (PCC) between experimentally measured kcat values and predicted kcat values of UniKP for wild type enzymes (a) (N = 936) and mutated enzymes (b) (N = 748). The color gradient represents the density of data points, ranging from blue (0.02) to red (0.28). c PCC values of wild-type and mutated enzymes on the test set of DLKcat and UniKP. Dark bars represent PCC values of DLKcat and light bars for UniKP. Source data are provided as a Source Data file.
Fig. 4 UniKP markedly discriminates kcat values of enzymes and their mutants. Scatter plot illustrating the Pearson coefficient correlation (PCC) between experimentally measured kcat values and predicted kcat values of UniKP for wild type enzymes (a) (N = 936) and mutated enzymes (b) (N = 748). The color gradient represents the density of data points, ranging from blue (0.02) to red (0.28). c PCC values of wild-type and mutated enzymes on the test set of DLKcat and UniKP. Dark bars represent PCC values of DLKcat and light bars for UniKP. Source data are provided as a Source Data file.CC BY · PubMed Central
Fig. 5 A two-layer framework considering environmental factors. a A two-layer framework called EF-UniKP that consists of a base layer and a meta layer. The base layer contains two models, namely UniKP and Revised UniKP. The UniKP takes the concatenated representation vector of the enzyme and substrate as input, while the Revised UniKP uses a concatenated representation vector of the enzyme and substrate, combined with the pH or temperature value. Both models are trained using the Extra Trees algorithm. The meta layer of this framework includes a linear regressor that uses the predicted kcat values from both the UniKP and Revised UniKP to predict the final kcat value. Scatter plot illustrating the Pearson coefficient correlation (PCC) between experimentally measured kcat values and predicted kcat values of Revised UniKP for pH set (b) (N = 636) and temperature set (c) (N = 572). The color gradient represents the density of data points, ranging from blue (0.02) to red (0.28). d Coefficient of determination (R2) values between experimentally measured kcat values and predicted kcat values on pH and temperature test sets of EF-UniKP, Revised UniKP and UniKP. Light bars represent R2 of EF-UniKP, dark bars for Revised UniKP and darkish bars for UniKP. e R2 values between experimentally measured kcat values and predicted kcat values on more strict pH and temperature test sets of EF-UniKP, Revised UniKP and UniKP. These are the samples in the test set where at least either the substrate or enzyme was not included in the training set, resulting in 62 and 61 samples for pH and temperature, respectively. Light bars represent R2 of EF-UniKP, dark bars for Revised UniKP and darkish bars for UniKP. Source data are provided as a Source Data file.
Fig. 5 A two-layer framework considering environmental factors. a A two-layer framework called EF-UniKP that consists of a base layer and a meta layer. The base layer contains two models, namely UniKP and Revised UniKP. The UniKP takes the concatenated representation vector of the enzyme and substrate as input, while the Revised UniKP uses a concatenated representation vector of the enzyme and substrate, combined with the pH or temperature value. Both models are trained using the Extra Trees algorithm. The meta layer of this framework includes a linear regressor that uses the predicted kcat values from both the UniKP and Revised UniKP to predict the final kcat value. Scatter plot illustrating the Pearson coefficient correlation (PCC) between experimentally measured kcat values and predicted kcat values of Revised UniKP for pH set (b) (N = 636) and temperature set (c) (N = 572). The color gradient represents the density of data points, ranging from blue (0.02) to red (0.28). d Coefficient of determination (R2) values between experimentally measured kcat values and predicted kcat values on pH and temperature test sets of EF-UniKP, Revised UniKP and UniKP. Light bars represent R2 of EF-UniKP, dark bars for Revised UniKP and darkish bars for UniKP. e R2 values between experimentally measured kcat values and predicted kcat values on more strict pH and temperature test sets of EF-UniKP, Revised UniKP and UniKP. These are the samples in the test set where at least either the substrate or enzyme was not included in the training set, resulting in 62 and 61 samples for pH and temperature, respectively. Light bars represent R2 of EF-UniKP, dark bars for Revised UniKP and darkish bars for UniKP. Source data are provided as a Source Data file.CC BY · PubMed Central
Fig. 6 Enhancing high kcat prediction through re-weighting methods and unified framework for Km and kcat / Km predictions. a The distribution of kcat values in the kcat dataset. All samples are divided into 50 bins. b The absolute error between experimentally measured kcat values and predicted kcat values of each sample. The kcat values of all samples were predicted independently using five-fold cross-validation. c Root mean square error (RMSE) between experimentally measured kcat values and predicted kcat values of 149 samples with kcat values higher than 4 (logarithm value) using various re-weighting methods and the initial UniKP. d, e RMSE, coefficient of determination (R2) between experimentally measured Km values and predicted Km values on Km test set. f Scatter plot illustrating the Pearson coefficient correlation (PCC) between experimentally measured kcat / Km values and predicted kcat / Km values of UniKP for kcat / Km dataset (N = 910). The color gradient represents the density of data points, ranging from blue (0.02) to red (0.28). Source data are provided as a Source Data file.
Fig. 6 Enhancing high kcat prediction through re-weighting methods and unified framework for Km and kcat / Km predictions. a The distribution of kcat values in the kcat dataset. All samples are divided into 50 bins. b The absolute error between experimentally measured kcat values and predicted kcat values of each sample. The kcat values of all samples were predicted independently using five-fold cross-validation. c Root mean square error (RMSE) between experimentally measured kcat values and predicted kcat values of 149 samples with kcat values higher than 4 (logarithm value) using various re-weighting methods and the initial UniKP. d, e RMSE, coefficient of determination (R2) between experimentally measured Km values and predicted Km values on Km test set. f Scatter plot illustrating the Pearson coefficient correlation (PCC) between experimentally measured kcat / Km values and predicted kcat / Km values of UniKP for kcat / Km dataset (N = 910). The color gradient represents the density of data points, ranging from blue (0.02) to red (0.28). Source data are provided as a Source Data file.CC BY · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Show all 54 references

Research works printed but not linked: 4

  1. Honda, S. et al. Smiles transformer: Pre-trained molecular fingerprint for low data drug discovery. arXiv preprint arXiv:1911.04738 (2019). · arXiv
  2. Van der Maaten, L. & Hinton, G. Visualizing data using t-SNE. J. Mach. Learning Res. 9, 11 (2008).
  3. Kroll, A. and Lercher, M. J. Machine learning models for the prediction of enzyme properties should be tested on proteins not used for model training. bioRxiv https://doi.org/10.1101/2023.02.06.526991 (2023). · DOI
  4. Li, Y. Deep reinforcement learning: An overview. arXiv preprint arXiv:1701.07274 (2017). · arXiv

Cited by

Sorted by citations

Related

See all

Metadata

Authors
Yu, H., Deng, H., He, J., Keasling, J. D., & Luo, X.
Affiliations
  • Chinese Academy of Sciences (CN)
  • Lawrence Berkeley National Laboratory (US)
Corresponding author
Luo, Xiaozhou
ORCID
Venue
Nature Communications
Publisher
Springer Science and Business Media LLC
Published
2023-12-11
Volume
14
Issue
1
Pages
8211-8211
ISSN
2041-1723 (ISSN-L)
DOI
10.1038/s41467-023-44113-1
PMC
PMC10713628
OpenAlex
W4389546039
Cited by
226
References
54
Citations by year
2026: 96 · 2025: 88 · 2024: 39 · 2023: 3
FWCI
20.23
Type
journal article
Language
en
Open access
Licence
Locations
  • Nature Communications (publishedVersion, cc-by)
  • Research Square (acceptedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • eScholarship (California Digital Library) (submittedVersion, cc-by)
  • OSTI OAI (U.S. Department of Energy Office of Scientific and Technical Information) (submittedVersion, cc-by)
  • Technical University of Denmark, DTU Orbit (Technical University of Denmark, DTU) (submittedVersion, other-oa)
  • RePEc: Research Papers in Economics (submittedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
National Natural Science Foundation of China (32071421)
Topics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Machine Learning in Bioinformatics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Enzyme Catalysis and Immobilization
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Protein Structure and Dynamics
Keywords
Michaelis constant, catalytic efficiency, protein sequences, substrate structure, pre-trained language models, environmental factors, pH and temperature effects, enzyme discovery, directed evolution, enzyme engineering
MeSH terms
Biotechnology, Catalysis, Kinetics, Temperature

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review