improve goodness of fit assessments #33
Assignees
Comments
|
#34 updates |
jbloom
added a commit
that referenced
this issue
Mar 26, 2024
- In `process_plate_curvefit_qc` in the YAML configuration, there is a new key called `goodness_of_fit` and now both `min_R2` (the minimum coefficient of determination) and `max_RMSD` (the maximum mean square deviation) for each curve fit are specified as keys under that. The curves are then filtered to retain only those that meet *either* of these criteria (so must fail both to be dropped). Addresses [this issue](#33) and [this issue](jbloomlab/neutcurve#55 (comment)). Alongside this change, the `rmsd` is now reported in key output files. Also, in the tabulation of failures, `fails_min_R2` now becomes `fails_goodness_of_fit`. - This is a **backward-incompatible change** in the configuration YAML. Previously `min_R2` was a standalone key under `process_plate_curvefit_qc`; now `goodness_of_fit` is the required key and `min_R2` and `max_RMSD` are required keys under it. - Added another plate (of H3N2 rather than H1N1) to the `test_example` to test some of the changes introduced in this version.
# for free
to join this conversation on GitHub.
Already have an account?
# to comment
Previously, the goodness of fit was assessed just using the coefficient of determination (
r2). This is fairly reasonable when the full sigmoid is fit.However, if the points fall mostly on a line (either never reaches neutralization or always neutralized at concentrations fit), then the coefficient of determination may be poor even if the data are fit well.
In
neutcurveversion 2.0.0, now the root mean square deviation (rmsd) of the fit is also reported (see here).The goodness of fit assessment QC should therefore now be updated to be something like this:
Either
r2) is close to one, say > 0.8rmsd) is close to zero, say < 0.05These new criteria need to be implemented into the pipeline.
The text was updated successfully, but these errors were encountered: