I'd tried to analyse my silent file that came from other concatenated silent files (cat *.out > 5novo.out) running the scoring application through my command line  but the program gave me this message:
core.io.pdb.file_data: [ WARNING ] discarding 3 atoms at position 420 in file 1IYT.pdb. Best match rsd_type: ALA_p:CtermProteinFull
core.pack.task: Packer task: initialize from command line()
protocols.evaluation.ChiWellRmsdEvaluatorCreator: Evaluation Creator active ...
protocols.jobdist.JobDistributors: Looking for an available job: 1 1 S_00000001 1
Starting work on structure: S_00000001_0001 <--- S_00000001
protocols.jobdist.main: Working on: S_00000001
ERROR: Illegal attempt to score with non-identical atom set between pose and etable
ERROR:: Exit from: src/core/scoring/etable/EtableEnergy.cc line: 104
I'd searched throughout this forum and I only find this post with similar (unsolved) problem (https://www.rosettacommons.org/node/2335). I know that my reference structure (1IYT) and my predicted structure (5novo.out) aren't identical (i.e. hasn't the same aminoacid sequences) and thus I decided to use the in:file:centroid option but the problem persists. I wonder to know if somebody could help me to circunvemt this error message.
 Command line
/home/guest/Downloads/rosetta3.5/rosetta_source/bin/score.linuxgccrelease -database /home/guest/Downloads/rosetta3.5/rosetta_database in:file:fullatom -in:file:silent 5novo.out -out:pdb -out:file:silent 5novo.1iyt.out -relax:fast -out:output -out:file:scorefile 5novo.1iyt.data -in:file:native 1IYT.pdb -silent_read_through_errors
That error message is what you would expect if you're scoring a fullatom structure with a centroid scorefunction (or vice versa).
Are all your input silent files of the same type? You may have some centroid-mode ones and some fullatom ones, which may lead to the issues regardless of the scorefunction used. If you're rescoring centroid mode files, you may also want to try explicitly setting the weights file to a centroid-mode scorefunction.
BTW, if having different lengths of proteins in a silent file is an issue for you (and I think it generally is), then reading with -in:file:centroid won't help any. Also, if it's just the native file which is of different length, you shouldn't need any special flags to read things in - native comparisons work with 1:1 mapping of residues starting with the first - any excess will be effectively ignored. If that doesn't work for you, you need to trim/pad out the native structure to match the length of your input structures.
Thanks rmoretti for your help, but I checked my input files (1IYT.pdb and 5novo.out) and they have the same numbers of aminoacids with a single mutation only at the 11th residue.
As my out file comes from ab initio relax I guess it is full atom since I used -out::file:silent bamil5_50k.out. Anyway, is there some other tips how I can analyse these data?
Outputting to a silent file doesn't necessarily mean a fullatom structure - both PDBs and silent files can hold both centroid and fullatom structures.
For PDBs, telling the difference is easy. If you have ATOM lines with "CEN" atoms, it's a centroid mode structure. If you have the standard sidechain atoms (besides CA & CB), then it's fullatom.
For silent files, it's a little bit harder. For binary format (where most of the content looks like letters and numbers in an unintelligible mass), if all the lines with the "gibberish" are the same length, then it's centroid mode (as all centroid mode protein residues have the same number of atoms). If they're different, it's fullatom.
For ASCII formatted silent files (with tabulated numeric results), the centroid mode files will have 6 columns of numeric values (9 columns total), whereas the fullatom ones will have 10 columns of numeric values (13 columns total).