lukasmueller
commented
5 months ago Repeat measurements can only be solved if we store additional metadata with each variable. Proposal: a new variable property, repeat_type, defines how the variable behaves for multiple measurements. Possible repeat_types could be single, multiple, and time_series. When single is associated with a trait, it will overwrite the last trait value. If multiple is associated with the trait, values will continue to be added, including time stamps. Averages can still be made over all the values, as they all represent the same phenotype. For time_series, multiple measurements are recorded, but the resulting data cannot be averaged, so on the trial detail page no average will be shown for them. It can be displayed as a growth curve, or used to derive growth parameters, etc.
Variables with multiple and time_series associations, when downloaded in spreadsheet format, could be stored in one cell separated by a delimiter, such as |. In the extended format, the measurement timestamp is given in parenthesis, so an example cell could contain 56(2024-01-20 12:00)|68(2024-01-21)|63(2024-01-27)
For multiple type variables, downloads could also average these numbers upon request for easier downstream analysis
hkmanching
Expected Behavior
If there are repeat measures for a specific observation unit/trait combo, then the phenotype download will return the oldest observation as determined by the collection date of the observation stored in the database. This is inconsistent with a user's expectation that the latest observation be returned in the file.
This overlaps with issue https://github.com/solgenomics/sgn/issues/3630
Steps to Reproduce