cxtrain command is not working

Hello,

I'm trying to calculate logD values based on my experimentally determined pKa values. I was told I could create a training library and I did so using Instant JChem. I exported the file as .sdf and now i'm stuck. According to the instructions on the ChemAxon website: 

The training library can be created using the cxtrain command line tool from an input structural file:

cxtrain pka -i [library name] [training file] 

What does it even mean? Where am I supposed to write this command? I've tried everywhere and still can't make it work. The only thing that runs scripts seems to be the Groovy script i can create from IJC, but again if I create a new script and write the command, it gives me an error message.

groovy.lang.MissingPropertyException: No such property: cxtrain for class: mypka_tach

    at mypka_tach(<self>:15)

Can anyone help with this? Maybe is something very obvious for people who do programming, but i really can't understand what to do.

Many thanks,

Luisa

Dear Luisa, 

The cxcalc is a command line tool, which means that the commands in the documentation should be run in the command line terminal on your operating system. The examples in the documentation are taken from the UNIX command line. Command line interfaces are available on any operating system. 

Does this help?

Daniel

Dear Daniel,

Thanks for your reply, but unfortunately it didn't help. I'm a chemist with no experience in programming, so what you wrote didn't make much sense to me. I'm on a Windows 8 system and i'm using Instant JChem version 15.4.13.0. I have tried to launch Instant JChem from Command Prompt and see if i could use the cxtrain command from there, but it doesn't work and it just open the user interface of the programme. 

Would you be so kind to go through a step-by-step of what i have to do on my windows system to have it working, please? 

I'd really appreciate your help,

Many thanks

Luisa

Dear Luisa, 

You have to do the following steps to run cxcalc in the command line prompt:

1. Search the command line tool on your system. On Win 8 the easiest way to do this is to type in the Start Menu cmd.     That will find your the prompt.

2. Search the cxcalc installation directory on your system. For me it is C:\Program Files (x86)\ChemAxon\JChem\bin.

3. Enter this directory. The cxcalc should be installed in this directory. 

4. From this directory you should be able to run cxcalc as you did it.

However, to run cxcalc for files this way you should also look up the file locations, otherwise cxcalc won't be able to find it. You can find this location in the Win file search. 

I hope this helps. 

Daniel

Dear Daniel,

Thank you very much for your reply, it was very kind of you to write all the instructions for me.

I had actually tried what you told me before sending you my last message and i've tried again now. In the InstantJChem\bin folder there's only the instantjchem.exe file, which i suppose is what we are looking for?

If i type cxcalc or cxtrain at this point, command prompt comes back with the message 'cxtrain' is not recognised as an internal or external command, operable program or batch file.

I've also tried to have InstantJChem starting from command prompt and then type cxcalc or cxtrain once the program was fully loaded, and again i had the same message in command prompt. What am I doing wrong?

Thanks for your time, sorry I'm being a bit of a pain!

Luisa

Dear Luisa, 

You should install either Instant JChem or MarvinBeans on your machine. As you only want to use the training functionality currently, I would suggest you download only the latest MarvinBeans from here:

https://www.chemaxon.com/download/marvin-suite/#mbeans

This will install the cxcalc and cxtrain command line tools in the /Program Files (x86)/ChemAxon/MarvinBeans/bin library. So after installation you should find these two tools (and others as well) in that library.

From there you can run the tools as I wrote yesterday. 

I would also point out again that you should provide cxcalc the absolute input file path in order to run it. If cxcalc doesn't find the file you want to use as an input, it won't be able to perform the calculation on it. So make sure it finds it in your library system.

I hope this helps. 

Daniel

 

Dear Daniel,

Thanks again for your help! Finally it works, so really thanks for your time.

I have another problem now, though! So, what I'm trying to do is to have Marvin Sketch calculating the logD values for my compounds from the experimental pKa values. To do so, I created the library with all my compounds and the experimental pKa values, which was what I was trying to sort out in my previous post.

The problem is that now, even when I apply the correction library, the pKa values that the program is giving me are considerably different (up to 2 pKa unit) from the ones I've put for that same compound in my library. The values differ whether I apply or not the correction library (which is a good sign), but as it stands I don't think I can trust the calculated logD values. As I'm trying to relate the activity of my library to the physical-chemical properties of each compound, you can understand that this is crucial. Is there any way to base the calculation for the logD on the actual experimental values for my compounds (and not on the estimated pKa, with or without correction)?

Thanks again,

Luisa 

Dear Luisa, 

Without having the molecule set you used for training it seems to us that the pKa training failed. To investigate further this issue first we have to know whether the training really failed or not. Can you see no difference between the pKa values before and after the training? If possible, can you send us some test set that actually produced bad results?

To answer your second question: currently there is no possibility to use measured pKa values directly for logD prediction. This can only be done via training. 

Best regards, 

Daniel

Dear Daniel, 

I think that the training library was set up successfully, because the pKa values for a given molecule are different according on whether i'm applying the training library or not. The problem is that for all of the molecules the estimated values are way out from the experimental ones even when I apply the correction library. If anything, the correction library actually made them worse! 

I have attached the .sdf file that I used for my library, so you can check if i've done something wrong.

Thanks

Luisa

Dear Luisa,

The fact that the pKa values before and after the training are different doesn't mean that training file was prepared correctly. You also have to make sure that the measured (experimental) pKa values were assigned to the correct atoms/atom indeces in the training file. 

I checked the first 3 molecules in your attached file for this, and they were correct. Can you please check all your molecules ?

Thank you, 

Daniel

Dear Daniel,

Thanks for your email. I've checked the atom numbering and my assignment in the file for all the molecules and they seem correct to me.

Cheers,

Luisa

Dear Luisa, 

Thanks for the checking. We are going to investigate this issue then, and get back to you soon.

Daniel

Dear Daniel,

Do you have any updates on my issue with the pKa training?

Thanks,

Luisa

Dear Luisa, 

Sorry for the late response. We investigated your issue again. Using your experimental data we are going to improve our pKa calculation. This bugfix will be available in the next week's release. So I recommend you to upgrade your software next week. After that please re-run your calculations.

I hope it is going to help.

Best regards, 

Daniel

Dear Luisa, 

We have one question about your test set: the 3 pKa values for molecule Nr. 9 in the test set seem to be very similar. Can you check whether they are really the correct experimental values?

Thanks, 

Daniel

Dear Daniel,

Thanks to you and the team for fixing the bug. I'll download the new version next week and i'll let you know how it goes. Will I have to re-train the library through the cxtrain command or is that going to still be operative after updating the version?

As for your question, the pKa values where measured through potentiometric titration and for compound 9 they were found to be 6.31, 6.09 and 5.69. I suppose the presence of the very electron-withdrawing CF3 group is affecting the amines, this is why they are all so similar.

Please let me know if there's anything else you need and thanks again for your help.

Luisa

Hi,

I'm having trouble with cxtrain also. I am in C:\Program Files (x86)\ChemAxon\MarvinBeans\bin and I can type in cxtrain -h.

But when I type in cxtrain pka -i mypka pKa_trainingset.sdf, the program says: "Exception in thread "main" java.io.FileNotFoundException: pKa_trainingset.sdf (The system cannot find the file specified)" and then some more code after.

I understand that I may need to "point" the system to the location of this file, which is in C:\Program Files (x86)\ChemAxon\MarvinBeans\help\calculations\training_files. How do I do this?

Thank you,

Ben

Hi, 

The reason for the exception is that you have to provide cxtrain a training file to run on. 

The best way to do this is to add to the PATH environment variable the directory of the cxtrain, so you can run it from everywhere. Then enter into the the directory where the training file is and run the library generation. 

I suggest you look up how to set the PATH variable on the internet. 

Daniel

Dear Luisa, 

We will provide a fix this week for your original problem. However, we still have things to check and fix related to your problem, so we need time to provide the final fix. 

Thank you for your patience, 

Daniel

Dear Daniel,

Thanks for the update.  Hope you'll be able to fix all of the problems soon.

Cheers,

Luisa