Introducing Parameter Clustering to the OED Procedure for Model Calibration of a Synthetic Inducible Promoter in S. cerevisiae

Round 1

Reviewer 1 Report

The paper describes the application of clustering methods to tackle the problem of calibrating the parameters of synthetic models.

I have some major concerns:

• The effectiveness of the methodology presented in this paper should be corroborated by comparing it with some standard approach for PE based on well known and widely used metaheuristics, such as CMA-ES and FST-PSO.
• There is an inconsistency between Fig 1 and Fig 2. The authors should explain why they only consider a subset of the model presented in Fig. 1 and better explain the ODEs in Eq. 1 and how they relate to Fig. 2
• The results in Section 3.1.1 must be better described and commented by the authors, as it is unclear the relevance of the outcomes of the clustering process
• It is unclear what the results presented in Section 3.1.2 demonstrate. The author should clarify this point
• Equation 8 is not properly described in the text
• I wonder what is obtained from the PE presented in the paper in terms of dynamics obtained with the estimated parameters compared to the experimental data used in the estimation process. The author should present and discuss such comparison
• Lastly, the most important observation is that English language must be carefully revised.

I also have some minor suggestions:

• the concepts of D- A- E-optimality must be described in the introduction
• at lines 79, 172, 180, 288, 306, replace "chapter" with "section"
• in Figure 1, the bounding of IPTG with LacI is represented as an inhibition arrow. I suggest to modify the representation
• after Eq 1, a ] is missing
• there are some typos throughout the text

Author Response

Dear reviewer,

First, thank you very much for your review. Please find the modified paper draft according to your comments (the modified parts are high-lighted in red), and here I am providing some feedbacks points by points.

For your major concerns:

• The effectiveness of the methodology presented in this paper should be corroborated by comparing it with some standard approach for PE based on well known and widely used metaheuristics, such as CMA-ES and FST-PSO.

Thank you for pointing out these alrogithms. References for these approaches have been added to section 2.5 and prospect parts. I would be very happy to try with these approaches in my next work (In fact, I have been using particle swarm optimization family in another project since last year).

The reason why they were not used is becase the work in this study is mainly carried out on a toolbox particularly developed for the PE/OED of biological networks (AMIGO2, ref 61), which does not have these algorithms on its NLP solver list (very likely because of the delay from computing field to biology field). For metaheuristic algorithms, it supports eSS, GLOBALm, SSm, and fSSm. Moreover (may be not very relevant), in a previous study (ref. 3), I once compared DE and eSS on some similar PE and OED tasks because I thought DE should works better than eSS, and find eSS is faster and more robust than DE on these tasks.

• There is an inconsistency between Fig 1 and Fig 2. The authors should explain why they only consider a subset of the model presented in Fig. 1 and better explain the ODEs in Eq. 1 and how they relate to Fig. 2

Explainations have been added to the script.

• The results in Section 3.1.1 must be better described and commented by the authors, as it is unclear the relevance of the outcomes of the clustering process

Explainations have been added to the script. The main messages want to deliver are:

1. The informativeness of random stimuli and shallow OEDs are significantly different, so only shallow OEDs are used to guide the parameter clustering;

2. The clustering results of the two different approaches give slightly different results, but they both reflect the inner property of the model.

• It is unclear what the results presented in Section 3.1.2 demonstrate. The author should clarify this point

Explainations have been added to the script. The main messages want to deliver are:

1. The clustering results vary with the initial parameter guesses.

2. The results varies but not completely random, the results still reflect the inner property of the model and connections between parameters.

• Equation 8 is not properly described in the text

Explainations have been added to the script.

• I wonder what is obtained from the PE presented in the paper in terms of dynamics obtained with the estimated parameters compared to the experimental data used in the estimation process. The author should present and discuss such comparison

I am afraid that I do have some difficulties in understanding what comparison you are suggesting exactly, because it seems like it is comparing the parameter estimation results to the experimental data (which is a bit confusing to me)?

If you are suggesting a plot about how the estimation accuracy increases after carrying out the 1st , 2nd, and 3rd sub-experiments, it needs some time (more than 1 week) to generate this data and I am doing it now. Please let me know if I get it right and I will add the plots and discussions in the next revision.

• Lastly, the most important observation is that English language must be carefully revised.

Totally agree. Although I have been checking the language by myself, I understand that there could be linguistic errors or typo. I am planing to use the mdpi language service once we are all happy with the content.

I also have some minor suggestions:

• the concepts of D- A- E-optimality must be described in the introduction

Details added.

• at lines 79, 172, 180, 288, 306, replace "chapter" with "section"

Words replaced.

• in Figure 1, the bounding of IPTG with LacI is represented as an inhibition arrow. I suggest to modify the representation

It is indeed an inhibition. IPTG indirectly promote the expression level by relieving the inhibition of LacI upon LacO (you can see an inhibition arrow from LacI to LacO). IPTG bounds to LacI and occupies its bounding site, so that there are less free LacI which can block the transcription of Citrine by bounding to LacO.

This mechanism is explained in ref.37, and this kind of indirect promotion mechanism can be found also in some other biological regulation systems such as the following example (relieving the autoinhibited Ephexin4)

*.Zhang, M., Lin, L., Wang, C., & Zhu, J. (2021). Double inhibition and activation mechanisms of Ephexin family RhoGEFs. Proceedings of the National Academy of Sciences, 118(8).

• after Eq 1, a ] is missing

I am not exactly sure about what this comment is about... I added explanations of the model parameters, which were not included.

• there are some typos throughout the text

I am planing to use the mdpi language service once we are all happy with the content.

Thank you again for your review.

Best wishes

Reviewer 2 Report

Review on ‘Parameter Clustering to Improve the OED Efficiency for Model Calibration of a Synthetic Inducible Promoter in S.cerevisiae’ 

 

This paper focus on synthetic gene circuits. The problem is interesting and may have practical interest in biological and pharmaceutical research and development. However, the paper does not allow one to evaluate the real novelty and applicability of the results obtained in it. Weakness of main contributions,The author should explain better how key advances in optimizing an experiment sequence for calibrating a model improve the calibration accuracy. Specific comments:Lines 108 -114: Eq. (2), matrix σ isn‘t defined. Why it is diagonal? What dimensions σ, y? Unclear indexation i,j in Eq (2). Unclear j, in . What procedure the parameters are evaluated (maybe MLE?).Line 115: should be referred to the literature for the computational complexity. This study is relevant to the mission of the journal Processes. I suggest not accepting the manuscript in the present form and returning the MS to the author for revision.

Author Response

Dear reviewer,

First, thank you very much for your review. Please find the modified paper draft according to your comments (modified parts are high-lighted), and here I am providing some feedbacks points by points.

This paper focus on synthetic gene circuits. The problem is interesting and may have practical interest in biological and pharmaceutical research and development. However, the paper does not allow one to evaluate the real novelty and applicability of the results obtained in it.

Thank you very much for your support, and I suppose you are suggesting to add a bit more background about what the pain point is and how well did previous studies tackle this problem (to an extent), and also how this paper compared to previous works.

For example, a "true" pain point of the current approaches is the high computational cost of FIM-based OED. It deters many biologists from adopting this process in their experiments in practices. I have added some more details about this in the introduction.

Moreover, I also make the comparison to previous works more clearier in the conclusion. (At the moment I tried not to add citations in the conclusion part, please let me know if you think it is better to have them in the last section).

Weakness of main contributions,The author should explain better how key advances in optimizing an experiment sequence for calibrating a model improve the calibration accuracy.

The main porpose is to introduce an apporach that significantly reduce the computational/time cost of the OED (so that OED can be carried out with acceptable cost for large models), whithout sacrificing the calibration accuracy. But in this study the mean accuracy indeed increases with the new approach, which is even better.

More information about the improvements in accuracy has been provided in the results and conclusions, please let me know if you think this way is better or still need to improve.

Specific comments:Lines 108 -114: Eq. (2), matrix σ isn‘t defined. Why it is diagonal? What dimensions σ, y? Unclear indexation i,j in Eq (2). Unclear j, in . What procedure the parameters are evaluated (maybe MLE?).

These details have been added to the script. BTW, the PE is carried out with the weighted least squares fitting.

Line 115: should be referred to the literature for the computational complexity. 

Definitely. I did a very brief search before, but did not find a very ideal reference. This time I checked more carefully and added a few references.

This study is relevant to the mission of the journal Processes. I suggest not accepting the manuscript in the present form and returning the MS to the author for revision.

Many thanks for your time and review again.

P.S. Although I have been checking the language by myself, I know that there could be some linguistic errors or typo. I am planing to use the mdpi language service once we are all happy with the content.

Best wishes

Round 2

Reviewer 1 Report

I am satisfied with the improvements of the manuscript. 

At this stage, I suggest to revise the English language before publication.