Hi @qiyifei1 , thanks for the bug report. Looks like PSF files can have insertion codes...

@qiyifei1 how did you generate this PSF file?

I have never heard of insertion codes in PSF files. Can you read this file in VMD?

It was created using CHARMM-GUI which uses CHARMM to generate psf files. VMD is OK (display and selection).
Here is an example
http://protein.org.cn/uploads/step1_pdbreader.psf

@qiyifei1 thanks for the example. If CHARMM-GUI produced the file then we should be able to parse it – but apparently we cannot, so this is a bug.

So has this bug been fixed? The psf topology file still cannot be loaded. The mdanalysis version of 2.6.1 is used.
The script I used is 'md = mda.Universe("../cg.psf", "./run.xtc")',but it failed and print as follows:
alueError Traceback (most recent call last)
File ~/software/miniconda3/envs/mdanalysis/lib/python3.10/site-packages/MDAnalysis/core/universe.py:110, in _topology_from_file_like(topology_file, topology_format, **kwargs)
109 with parser(topology_file) as p:
--> 110 topology = p.parse(**kwargs)
111 except (IOError, OSError) as err:
112 # There are 2 kinds of errors that might be raised here:
113 # one because the file isn't present
114 # or the permissions are bad, second when the parser fails

File ~/software/miniconda3/envs/mdanalysis/lib/python3.10/site-packages/MDAnalysis/topology/PSFParser.py:149, in PSFParser.parse(self, **kwargs)
147 next(psffile)
148 top.add_TopologyAttr(
--> 149 attr(self._parse_sec(psffile, info)))
150 except StopIteration:
151 # Reached the end of the file before we expected

File ~/software/miniconda3/envs/mdanalysis/lib/python3.10/site-packages/MDAnalysis/topology/PSFParser.py:178, in PSFParser._parse_sec(self, psffile, section_info)
177 logger.error(err)
--> 178 raise ValueError(err)
179 # Now figure out how many lines to read

ValueError: Expected section NTHETA but found 12

During handling of the above exception, another exception occurred:
...
128 "Error: {2}".format(topology_file, parser, err))
129 return topology

ValueError: Failed to construct topology from file ../cg.psf with parser <class 'MDAnalysis.topology.PSFParser.PSFParser'>.
Error: Expected section NTHETA but found 12

There are some custom residues in this psf file.However,I can load this psf file by using VMD.Is there any other solution?
Kind regards.

@ldx022 does your PSF file contain insertion codes, as in the original report, i.e. residue numbers with directly a character following?

Thank you for your reply. For the psf file type, I only know that it is one of the topological formats.I'm not sure what insert code is, if I follow my own understanding, I have checked and there is no similar situation to the above 185A GLU.This psf file was generated using a coarse-grained program we developed. As a result, it has some custom residue types and atom types.Using vmd to load this file without any problem, I now attach the psf file(Because github upload has format restrictions, so I replaced the psf suffix with txt suffix)
cg.txt

I recently ran into the same issue where I was unable to parse PSF files generated by psfgen in VMD.

After checking the PSF format, surprise surprise resid is a string! see the format below

https://github.com/MDAnalysis/mdanalysis/blob/d366921b0f614f455f5cc90d4024b8a5817aa9ed/package/MDAnalysis/topology/PSFParser.py#L212C26-L212C29

In VMD, resid is read as a string and then converted to an integer using the atoi function.

I believe the optimal approach would be to maintain resid as an integer in MDAnalysis while adding an attribute residstr. Maybe a good GSOC starter?

Can someone dig out the PSF specs from the CHARMM website or the source?

If it’s clear that insertion codes are defined then we can parse them separately and have an attribute “icode” do residues, IIRC we have that for PDB.

I’d rather parse the correct semantics than having a residstr attribute.

From the latest charmm/source/io/psfres.F90 (fmt01 for no chemq; fmt01 for chemq; fmt01b for drude):

     # II,LSEGID,LRESID,LRES,TYPE(I),IAC(I),CG(I),AMASS(I),IMOVE(I),ECH(I),EHA(I)
     # PSF EXT
     fmt01= '(I10,1X,A8,1X,A8, 1X,A8,1X,A8,1X,I4,1X,2G14.6,I8)'
     fmt01a='(I10,1X,A8,1X,A8,1X,A8,1X,A8,1X,I4,1X,2G14.6,I8,2G14.6)'
     fmt01b='(I10,1X,A8,1X,A8,1X,A8,1X,A8,1X,I4,1X,2G14.6,I8,2G14.6,L1)'
     # PSF
     fmt01='(I8,1X,A4,1X,A4,1X,A4,1X,A4,1X,I4,1X,2G14.6,I8)'
     fmt01a='(I8,1X,A4,1X,A4,1X,A4,1X,A4,1X,I4,1X,2G14.6,I8,2G14.6)'
     fmt02='(I8,1X,A4,1X,A4,1X,A4,1X,A4,1X,A4,1X,2G14.6,I8)'
     fmt02a='(I8,1X,A4,1X,A4,1X,A4,1X,A4,1X,A4,1X,2G14.6,I8,2G14.6)'

LRESID -> A8 / A4

I believe CHARMM formats the resid to be left-justified and adds the insertion code at the end if present.

However, in the file generated by psfgen, the resid changes to hex (when it exceeds 99999?). It makes it difficult to differentiate the resid from the insertion code under these circumstances.

We can certainly be courteous and consider all conditions (if possible), or simply remove all non-numeric characters from the resid and parse the remainder (and adding 'residstr' as an attribute or not).