Assessment tools with "False" and "True" strings are parsed as boolean, breaking things

[surchs]

When I have an assessment tool column with unique values that look like the python special words for boolean values (i.e. "True" and "False"), then pandas reads the whole column as type boolean. If the "False" value is meant to indicate a MissingValue, then this is ignored now, because in the data dictionary it is described as a string, but in the loaded bagel.tsv, it is now a bool.

Here is an example of a data dictionary for a column with "True" and "False" values

  "moca_total_status": {
    "Description": "Montreal Cognitive Assessment",
    "Levels": {
      "True": "Completed",
      "False": "Not completed"
    },
    "Annotations": {
      "IsAbout": {
        "TermURL": "bg:Assessment",
        "Label": "Assessment Tool"
      },
      "IsPartOf": {
        "TermURL": "cogAtlas:trm_57964b8a66aed",
        "Label": "Montreal Cognitive Assessment"
      },
      "MissingValues": ["False"]
    }
  }

and here is an example of what the corresponding bagel.tsv file would look like

participant_id	moca_total_status
sub-01	True
sub-02	False
sub-03	True

Then all three subjects (sub-01, sub-02, sub-03) would be shown to "have" the MOCA tool, because none of them would have a "False" value inside of the loaded bagel.tsv (because it got cast into a boolean and False != "False" in python).

Sneakily, all of this happens completely silently and without any error, because we are not checking whether the values described in the data dictionary actually exist in the loaded bagel.tsv file.

I think two things should be done:

• force bagel.tsv to be loaded as string type to avoid unwanted type casting #108
• ensure all discrete values exist in the data dictionary for bagel pheno #109