Using LOGP as P-Val

[C-Burden]

• PheWeb 1.3.16
• I am running Pheweb Process for a single file, and it results in an assertion error since I am using LOGP
• config.py:

hg_build_number = 38
pval_is_neglog10 = True
field_aliases = {'LOGP': 'pval'}

• Error Message:

Traceback (most recent call last):
  File "/app001/.local/lib/python3.9/site-packages/pheweb/load/parse_input_files.py", line 70, in convert
    writer.write_all(variants)
  File "/app001/.local/lib/python3.9/site-packages/pheweb/file_utils.py", line 378, in write_all
    for v in variants:
  File "/app001/.local/lib/python3.9/site-packages/pheweb/load/read_input_file.py", line 26, in get_variants
    yield from self._order_refalt_lexicographically(
  File "/app001/.local/lib/python3.9/site-packages/pheweb/load/read_input_file.py", line 59, in _order_refalt_lexicographically
    for v in sorted(tied_variants, key=lambda v:(v['ref'], v['alt'])):
  File "/app001/.local/lib/python3.9/site-packages/pheweb/load/read_input_file.py", line 141, in get_variants
    variant = self._parse_variant(values, colnames, colidx_for_field)
  File "/app001/.local/lib/python3.9/site-packages/pheweb/load/read_input_file.py", line 201, in _parse_variant
    raise PheWebError(
pheweb.utils.PheWebError: failed on field 'pval' attempting to convert value '1.00752' to type <class 'float'> with constraints {'aliases': ['PVALUE', 'P', 'P.VALUE'], 'required': True, 'type': <class 'float'>, 'nullable': True, 'range': [0, 1], 'sigfigs': 2, 'tooltip_lztemplate': {'condition': False, 'template': '{{#if pvalue|is_numeric}}P-value: <strong>{{pvalue|scinotation}}</strong><br>{{/if}}\n{{#if pval|is_numeric}}P-value: <strong>{{pval|scinotation}}</strong><br>{{/if}}'}, 'display': 'P-value', 'from_assoc_files': True} in '/app001/data/008.tsv.gz' on line with values ['1', '783125', 'A', 'G', '0.035415', '85589', '0.128243', '0.0775708', '1.00752'] given colnames ['chrom', 'pos', 'ref', 'alt', 'maf', 'n', 'beta', 'se', 'logp'] and field mapping {'chrom': 0, 'pos': 1, 'ref': 2, 'alt': 3, 'maf': 4, 'beta': 6, 'sebeta': 7, 'pval': 8}

• Sample of the GWAS file being used
  008.tsv

[ttbek]

I just ran into the same issue. I think you're just going to have to convert them and have zeros (which it will plot at an 'infinity for the -log10(0)') if you have extreme p-values. Even if the assertion would let it through, I think the approach in the code is to convert it back to a normal p-value anyway and then convert back again in the plotting:

        # It's POSSIBLE that conversion of -log10 representation is required
        if 'could_be_neglog10' in self._d and self._d['could_be_neglog10'] and conf.pval_is_neglog10():
            x = 10**-x

A more numerically stable approach would be for the plotting code to be rewritten assuming it is directly getting the minus log10 values and convert to those as normal if given regular p-values. I'm not sure what functions Python has available, but I was converting chi square values to p-values... and getting 0, but in R the calculation can be done logging both sides in the calculation to get a meaningful log(p-value) that you can do a change of log transform on.
common_all[which(common_all$PVAL == 0),]$minuslog10pval <- -pchisq(q = common_all[which(common_all$PVAL == 0),]$CHISQ, df = 1, lower.tail = FALSE, log = TRUE)/log(10)

I think Peter has mostly moved on from this project now and suggests that people use Pheweb2 https://github.com/GaglianoTaliun-Lab/PheWeb2
Personally... I feel like PheWeb2 complicates things significantly and introduces more outside dependencies... Maybe it will grow on me, but right now I don't like it.

Long story short, either use normal p-values or try Pheweb2 I think.