Dear Ekaterina,

You are right to point out that collaborative work might be impacted by these biases and one should take them into account. This could be used advantageously (instead of nefariously) as a way to systematically reproduce results on multiple systems: if the results converge then it increases the reliability of the result.

As I said earlier, this is a generalized problem of computer science, or rather software science, and is already well known in the domain. The paper I cited earlier merely highlighted the issue to the neuroscientific community, but these observations can be extended to SPM and CONN. If someday someone finds out a way to avoid this kind of bias, you can be sure it will be a revolution in the whole computer science field.

About using CONN on a cluster, first you can run it from scripts, you don't need to use the GUI. You can checkout the conn_batch.m and conn_modules.m files which provide APIs to run CONN entirely from the commandline or scripts, allowing batch scripting and to run on a non-GUI environment such as a cluster.

Secondly, you are right that there can be a licensing issue, but if you own one MATLAB license, you can build standalone CONN and SPM packages, which will then be runnable on a cluster without necessitating any MATLAB install. To do this, you can checkout the folder "standalone" inside the CONN folder, and a similar thing exists for SPM too. Although note that I did not test this yet by myself, and I am not entirely sure whether standalone CONN can use a standalone SPM, although this should be possible with slight modifications to the code in any case (if you try out, Alfonso can surely help!).

Hope this can help, good luck with your experiments!
Best regards,
Stephen