How can I estimate PD parameters for two - compartment model?

I have a two compartment model with two types of dosing: PK dose and PD dose (scheduled).

I have mapped the data correctly, but my estimate is far off from the reference paper I was using as a prototype. I know thet fit program is for parameter estimate, however I dont know how to assign the baseline dose. Note, the model has no variant.

I do not know what excatly I am doing wrong, and i really do need help with this; been stuck with it for more than a week.

[Arthur Goldsipe]

It's very difficult to figure out the problem without more details. Can you share the project you created and the results you are trying to reproduce?

[BLESSING AKAH]

The drug potency (k2), transient rate (k1) and initial tumor growth (wo) were estimated to be 13.8, 0.130, 0.0386 respectively.

I have tried all I could to get same paramter estimate using the build in model, but my estmate were all far from the actual estimate.

I do not know what exactly I am not doing well.

Please, I will apprecaite any help on how to get the exact estimated values given above.

Thanks,

Blessing.

Answer 1

[Arthur Goldsipe]

I looked at the project you attached in the comments. I didn't see any obvious problems. Your parameter estimates are within a factor of 2 of the ones you're trying to match. Do you have any reason to believe you should be able to reproduce the reported values more closely? If so, you will likely need to get more details from the author to make sure you are reproducing their model and their estimation methods. In fact, before you try to reproduce their estimation results, I recommend making sure you can reproduce their simulation results for a specific condition. For example, do the authors provide simulation results of tumor weight at the estimated parameter values? If so, can you simulate the same conditions and compare your simulation results? If those don't agree, then you may have a problem with your model. If those do agree, then I recommend double-checking that you're using the correct data and the same fitting approach (for example, the same error model).

Answer 2

[BLESSING AKAH]

Hi Arthur, thanks for the detailed information.

Yes, the author did provide the fit results for the estimated parameters: w0, k1 and k2. The initial values used by the author for the estimates were 0.04, 0.5 and 10 for the w0, k1, k2 respectively. The error model used by the author was exponential error model. I did replicate the model, but my estimates didn’t match with the values given by the author. Even when I tried estimating using the author built in model, I was unable to get the estimates for the parameters. I don’t know what I’m doing wrong. PS: I am currently using the author model as a reference on my project. Hence, I needed to get the exact parameter estimates to continue with my project. I am new to using sim biology, hence might be doing something wrong. Please, would you mind running the fit program with the same error model and initial parameters and see if you could get the exact values given by the author. I have been stuck on it for the past few weeks, and I really need my way out of it.

Please let me know if you need more clarification on that. Thanks so much for your help.

Best, Blessing.

[Arthur Goldsipe]

I wouldn't assume you've done anything wrong. It's very hard to reproduce the results in a publication unless the authors share complete details. This is especially important when you're trying to estimate parameters. Sometimes you will see that changing your parameter values does not singificantly affect the quality of the fit, so perhaps your estimated parameters are really just as good as the ones reported in the paper.

When I was in a similar situation, I contacted the authors and asked for code to reproduce their results. Many authors are happy to share that information. So if it's critical that you exactly reproduce their results, I suggest reaching out to the authors.

By the way, what's the paper? I can also see if anyone on my team is familiar with the paper and model.

[BLESSING AKAH]

Hi Arthur, thanks for your response.

The author shared the complete MATLAB file; the file I shared with you yesterday is from the author.

I did estimated the PK parameters correctly, but having problem with the PD estimates.

Quick question: for this model, do I need to add baseline dose in the fit program, like the PK dose and /or PD dose?

Also, the appropriate stats to log while estimating paramters, PK and/or PD parameters.

Thanks one again for your help.

Best,

Blessing.

[Arthur Goldsipe]

I took another look. I still don't think you're doing anything wrong. The authors did not provide enough information to replicate their estimation approach. In fact, I worry that their approach might not even be reproducible. The description in the paper is what I'd call "hand wavy." I'd summarize it as "we manually adjusted the parameters until some numbers and plots looked good."

And to answer your other questions, I think the relevant dosing is contained in the data, so I don't see any need for baseline dosing. And changing which states are logged should not affect the results of parameter estimation.

I also saw that the authors used version R2019a of MATLAB and SimBiology, so I opened their projects in that version. Even in that version, I could not reproduce the results they reported in the paper when running the fit programs in their projects. I also noticed that the authors' project files do not include any results from running fit programs. This also suggests to me that the reported parameter estimates were NOT obtained by running fit programs in SimBiology.

[BLESSING AKAH]

Hi Arthur, thanks again for your reply.

I had the same though as you did "the estiamtes isn't reproducible using the authors method. Also, I had thought that no need for baseline since the PD data already has the dose schedule in it.

Like i said earlier, I am new into Simbiology, in leiu of that I need to be assured i did followed the right step for the fitting program. Would you mind sharing your estimated value using exponential and constant error model, so I can compare with my estimated values.

Sorry for the lengthy question. I am grateful for your help.

Best,

Blessing.

[Arthur Goldsipe]

Hi,

My parameter estimates for the exponential error model are w0=0.0370, k1=0.0548, and k2=9.8666.

My parameter estimates for the constant error model are w0=0.0383, k1=0.1825, and k2=11.8121.

I'm attaching my project, which I created in R2021b, by adding two programs (Fit Data_exponential_error and Fit Data_constant_error) to the project you shared.

-Arthur

[BLESSING AKAH]

Hi Arthur,

I did went through the project you saved. The constant error you have matches with my estimates. However, the exponential error you saved, the values you mentioned that you got did not match with the file you saved.

The values you mentioned that you estimated were actually estimated using combined error model not exponential as you said.

The value from my exponential were, w0 = 0.0382, k1 = 339.26, k2 = 17.2548.

My thought is that the error model used by the authors were not exponential error. Hence, I can assert that my estimates are correct.

Also, I noticed the data only has PD data, no PK data (dose concentration). Are'nt PK data supposed to be added as baseline dose? Please, could you clarify this.

Best,

Blessing.

[Arthur Goldsipe]

Oh, glad I attached the project so that you could catch my mistake with the error model.

I'm sorry, but I don't have time to read the paper closely enough to figure out whether there's some missing data or dosing information. I suggest contacting the authors, since they would know best. I see contact information listed at the end of the paper.

If you figure out what's going on, I'd love to hear more about it. Feel free to contact me directly.

[BLESSING AKAH]

Thanks Arthur for your response; it does help solve my issue with the estimate. I greatly appreciate the consistency in your response. Sure, I will contact you directly if need be.