Deep learning allows genome-scale prediction of Michaelis constants from structural features

The Michaelis constant KM describes the affinity of an enzyme for a specific substrate and is a central parameter in studies of enzyme kinetics and cellular physiology. As measurements of KM are often difficult and time-consuming, experimental estimates exist for only a minority of enzyme–substrate combinations even in model organisms. Here, we build and train an organism-independent model that successfully predicts KM values for natural enzyme–substrate combinations using machine and deep learning methods. Predictions are based on a task-specific molecular fingerprint of the substrate, generated using a graph neural network, and on a deep numerical representation of the enzyme’s amino acid sequence. We provide genome-scale KM predictions for 47 model organisms, which can be used to approximately relate metabolite concentrations to cellular physiology and to aid in the parameterization of kinetic models of cellular metabolism.


In the manuscript "Prediction of Michaelis constants from structural features using deep learning" by Kroll et. al. a graph-based method of predicting Km for a ligand is illustrated utilizing both chemical descriptors as well as a binary vector of a protein domains. Broadly, the work done is executed well but conclusions are significantly overstated.
Response: We agree that limitations of our methodology were not emphasized enough in the previous version.
Action : We have addressed this point in two ways: we have improved the methodology by representing enzymes through UniRep vectors (deep learning representations of the amino acid sequences; l. 181), and we now clearly discuss the limitations of our approach (e.g., l. 332).

Response:
We thank the Reviewer for pointing out this omission.

Action:
We now discuss drug target binding affinity (DTBA) predictions in the introduction (l. 45) and compare our prediction quality with that of two DTBA models (l. 308).
4. Therefore, while this article is interesting and contributes to the field, it should do a better job placing the advance within the field, make more accurate claims about the utility and advance, and is likely more appropriate for a trade journal. I do apologize if this review seems overly harsh, however I was truly excited when I read the title and abstract and found myself deeply frustrated by the time I had concluded reading the article.

Response:
We apologize for having caused this frustration, and we hope that the Reviewer will find the revised manuscript somewhat less disappointing.
Action: See our responses to points 1-3.

Response:
We thank the reviewer for this positive assessment.

The authors present the model as the main result. However, to most biologists including me, the predicted values of Km are probably most valuable. I thus suggest the authors apply the model to all possible reactions in popular organisms such as human and mouse. Ideally, the results can be made available in the format of website (which can be updated as the model is updated in future). At least an excel file containing the numbers should be included in the supplementary materials.
Response: We agree that providing predictions for popular model organisms is indeed desirable.
Action: In Dataset S1, we now provide KM predictions for all reactions in 47 curated genome-scale metabolic models, including those for human, mouse, the yeast Saccharomyces cerevisiae, and E. coli.
2. The representation of the enzyme structure seems over-simplified. The author discussed a number of enzyme properties that can be considered when they become more widely available in future. The order of the domains (along the protein sequence) however is available. I wonder if this structurally important information can be incorporated into the model?

Response:
We thank the Reviewer for this suggestion. However, in light of this comment and comments by the other Reviewers, we feel that such a change would still not allow us to adequately represent enzyme amino acid sequences.

Action:
In the revised manuscript, we represent enzymes through a deep learning representation of the amino acid sequence, UniRep (Alley et al. 2019). This now allows us to make specific predictions for different sequences belonging to the same enzyme family, and leads to a substantial improvement of the predictions.

Can the authors discuss the applicability of the model to the prediction of Kcat?
Response: We agree that predicting kcat may be possible with a similar framework; an important difference would be the requirement to represent all substrates (and possibly products) of the enzyme as simultaneous inputs. While we find this an interesting suggestion, we consider it to be beyond the scope of our analysis and leave it for future work.

Action:
We now indicate the applicability of a similar framework to the prediction of kcat in the Discussion (l. 355).
4. The origins of the Y-axis of the MSE and R2 figures were randomly chosen at the moment. I suggest set all of them to 0.
Response: While we agree with the reviewer that setting all origins to zero would make it easier to visually compare across figures, the differences between methods in some of the figures are so small that they would be hard to distinguish.

The manuscript by Kroll et al. uses a combination of machine and deep learning to
predict Km values based on available information on substrate fingerprints. Currently, machine and deep learning are very "hot" topics in all fields. Their contribution to parameter prediction is of particular value to the systems biology community as it offers significant assistance in model parametrization.
The framework seems to be well researched and structured and the presentation of the paper is very good. The results also look very promising for Km prediction and it would indeed be very interesting to see results for Kcat parameters in the future. Also the github page with the model information is very practical and very well detailed.
A few points should be addressed before acceptance.

Response:
We thank the reviewer for this positive assessment.

The test procedure seems a little basic as there is only one test dataset. A full kfold cross validation would be better.
Response: We agree that performing a k-fold cross validation is a more common approach and provides additional information.

It would also be nice to get some empirical investigation as to how robust the results are to different choices of train/validation/test data. It would be interesting to see how much the hyperparameters change, whether the errors are stable etc.
Both of these issues could be addressed fairly simply by re-running the analysis on different splits.

Response:
We agree that such additional information is valuable.

Action:
The 5-fold cross validations we now perform demonstrate that the performance is relatively robust across different splits (Figures 2-4). Response: We thank the reviewer for this positive assessment.

(a) Regarding the training/test data, the authors use BRENDA which quite a standard database. I however don't see the data being available as supplementary. I understand that there is a script in github to do the preprocessing but since BRENDA can be updated I would make the data preprocessed available as a supplementary for better reproducibility and future comparisons. This would also be of help to the noncomputational readers of PLoS Biology. (b) I was a bit surprised to see that the authors don't use also SABIO-RK database, that seems quite a standard in the field. On a quick search they don't seem to include exactly the same information so I wonder why the authors did not use it for model building or testing (see comment below). (c) Finally, I was a bit surprised why the authors don't doo 5-or 10-fold cross validation although I have more questions regarding model testing below.
present in both training and test sets. That would be the fair approach to testing it in my opinion.
Response: As we now use data for individual enzyme-substrate combinations (rather than pooling the data for a given combination of protein domains), these numbers have changed. There are now 664 enzyme-substrate combinations in the test set that do not occur in the training set. For 57 enzyme-substrate combinations in the test set, neither enzyme nor substrate are represented in the training set.

Action:
The updated analysis of enzymes and substrates not represented in the training set is found in the section starting l. 228.
7. I appreciate the authors discuss the MSE differences with previous approaches despite not being possible to do a comprehensive comparison. I wonder why they don't discuss R2 too and there are a couple of previous works that are also not included in this discussion. I think this part should be elaborated more.

Response:
The previous work that we are aware of does not report R 2 values that can be compared to our work.

Action:
We now additionally compare the performance of our model to two studies predicting drug target binding affinities (DTBA), which report rm 2 values (l. 308). enzyme that explain experimental observations. This is just a suggestion but ending the paper with an example of use would make if far less dry and more palatable for non-expert experimental scientists.

Response:
We agree that providing KM estimates for a broad range of enzymes and model organisms is indeed desirable. While providing a shiny app is beyond the scope of our current study, providing predictions for a range of model organisms is indeed possible. We like the idea of supplying a simple use case at the end of the manuscript, which would indeed make it a more enjoyable read especially for many experimental scientists. However, while such use cases undoubtedly exist, we failed to identify an example that is simultaneously simple and illustrative.
Action: As suggested, we now provide KM predictions for all reactions in 47 curated genome-scale metabolic models, including those for human, mouse, Saccharomyces cerevisiae, and E. coli (Dataset S1).
10. In summary, I have enjoyed reading the manuscript and the article is clearly a step forward in Km prediction. I therefore strongly encourage the authors to perform the major revisions suggested or transfer the paper to a computational journal with fewer revisions needed.