Good strategy.
A good compression recipe.

The basic strategy the script uses:

* Finds the 'center' of the chain (using an algorithm by ssrsmith)
* Creates bands between each hydrophobic segment and the 'center' (as best it can given the script functions that are available)
* Starts the sequence listed below
** Gradually increases the band strength until there is movement of at least 10 points
** Disables bands and wiggles and shakes until the change is no more than .03 points
** Wiggles sidechains
** Backs off the band strength by .03
** Repeats enabling and disabling until there are at least 7 passes without any increase in score
* Does one final wiggle sidechains
* Restores the structure to the best score since the script started
* Quits, showing you the increase in score in the output

If you show output, it tells you exactly what it is doing. You should normally let the script run completely through to completion. It will, at worst, leave you right where you started, but normally get you at least a few points, and sometimes several hundred.

All of the significant numbers are passed into the Main routine at the end of the script. You can change those numbers to other values to get different effects. The values I have there are a compromise between getting the absolute best score and having the script run fast.

The two most important parameters:

If you reduce sigmaStableScore to .01, for example, the script will milk every increase in score it can when it wiggles and shakes, but will take significantly longer to run.

If you increase changeScore, it will create stronger bands causing a more dramatic change in the structure on each compression. This can be good or bad, depending on whether a dramatic change will find a lower energy state than the one you are in now.

Do you think a variant of this script could help to turn hydrophobic segments inside by rebuild?

If we could reliably determine which segments were hydrophobic.

It would be a different script, though.