Fragment generation is two steps. The hard-for-users step is secondary structure prediction generation, mostly because it relies on half a dozen bits of scattered software. The actual fragment generation is not that hard – if you already have SS predictions in hand, it’s pretty fast with the demo included in 3.3.
Errors in the input sequence (point mutations) are only relevant insofar as they affect the secondary structure prediction. If your mutation is going to turn a loop into a loop, it won’t matter. If your mutation is going to turn a helix-preferring region into a loop-preferring region, it can matter.
Errors in the _length_ of the sequence are hugely important. That will produce the equivalent of a frameshift mutation in the fragment file – all the fragments will apply to the wrong frames. You should regenerate fragments for indels. Rosetta will probably refuse to accept a fragment file with a different length from the protein.
If you have the SS predictions for the old sequence, you can manually edit those to include your indel (if it’s in a loop, just assign it the SS preference of its neighbors), then rerun only the second step of fragment picking. Or just let Robetta do it.
.
