Handling of p-values next steps

[tudorgroza]

As per: https://github.com/EBISPOT/goci/issues/210

B. Choose P < 1E-5 as definitive threshold and move to storing actual p-values to multiple decimal places as presented by authors

Requirements:

• change template validation from P ≤ 1E-5 to P < 1E-5, i.e. p-values of exactly 1E-5 should not pass validation
• restructure how p-values are stored in the database, to allow storage of multiple decimal places
• change curation interface/database validation to allow associations with 9.5E-6 ≤ P < 1E-5
• UI changes to the curation interface (and probably Search UI, download etc.) to display p-values with multiple decimal places

Pros:

• fits our published p-value threshold, "intuitive" number used as a cutoff
• include associations in the grey area of 9.5E-6 ≤ P < 1E-5, which we have "advertised" as eligible but are not currently including
• by not rounding, we retain more information and greater specificity about p-values
• imposes a hard automated threshold, not dependent on curators rounding or manually excluding any values

Cons:

• more significant changes to the database and curation interface
• inconsistency with older studies: associations in the grey area not included in older studies, now included under this model
• inconsistency with older studies: associations from older studies would continue to be stored as integer Mantissa and Exponent, while associations from newer studies would be stored as a float (either a single number, or as float Mantissa and integer Exponent) - potential problems when displaying them side by side in Search UI etc., e.g. if we convert old associations to float, then we might add erroneous specificity (e.g. 8E-6 > 8.00E-6 for an older study vs. 8.34E-6 for a newer study)?

[ljwh2]

See gwas-sumstats-validator#15 - it may be necessary to keep using mantissa & exponent in order to represent v. small p-values

@eks-ebi I think requirement point 3 is already implemented, please confirm.

[eks-ebi]

Yes, point 3 is implemented now.

[ljwh2]

@eks-ebi @earlEBI is the template validation part also implemented? ie

• change template validation from P ≤ 1E-5 to P < 1E-5, i.e. p-values of exactly 1E-5 should not pass validation

[eks-ebi]

Yes, I think so.

[ljwh2]

OK, moving to icebox for now since we won't make major changes to the data representation in the near future, and seem like there are other complications in handling very small numbers.