Unexpected behavior for binned continuous metadata

[kimandrews]

Two issues related to binned continuous metadata:

1. Frequencies panel shows bins based on "bounds" rather than"display"

Here is an example tb dataset generated using bins defined for the Number of drug resistant variants in the auspice-config.json like this:

   {
    "key": "num_dr_variants",
    "title": "Number of drug resistance variants",
    "type": "continuous",
    "legend": [
      { "value": 0,    "display": "0",     "bounds": [0, 0] },
      { "value": 1.5,  "display": "1-2",   "bounds": [0.5, 2.5] },
      { "value": 3.5,  "display": "3-4",   "bounds": [2.5, 4.5] },
      { "value": 5.5,  "display": "5-6",   "bounds": [4.5, 6.5] },
      { "value": 7.5,  "display": "7-8",   "bounds": [6.5, 8.5] },
      { "value": 9.5,  "display": "9-10",  "bounds": [8.5, 10.5] },
      { "value": 11.5, "display": "11+",   "bounds": [10.5, 100] }
    ]
  },

In the example dataset, the auspice frequencies panel shows the upper and lower values of "bounds" for each bin, which means that samples with 1-2 mutations are shown as 0.50-2.50, whereas it would be preferable to show the "display" value of 1-2. Changing the "bounds" to [1, 2] does not solve the problem because samples with 1 mutation do not fit into those bounds.

2. Coloring behavior seems to vary depending on whether samples are present within each bin

Here are 3 example datasets that differ in the number of samples, and whether the auspice_config.json has a "scale" section for the Percent reads mapped (note that all datasets have Number of drug resistant variants bins as described above, with no "scale" section):

• 8 samples, with "scale" section: https://nextstrain.org/staging/tb/n7/test1
• 8 samples, without "scale" section: https://nextstrain.org/staging/tb/n7/test2
• ~1000 samples, with "scale" section: https://nextstrain.org/staging/tb/aws/test/n1000c

Here is the 'Percent reads mapped' block WITH the "scale" section in the auspice_config.json:

{
    "key": "pct_reads_mapped",
    "title": "Percent reads mapped",
    "type": "continuous",
    "scale": [
      [80, "#DB2823"],
      [85, "#E1A13A"],
      [90, "#89BB6B"],
      [95, "#447DCC"],
      [100, "#5E1D9D"]
    ],
    "legend": [
      {"value": 82.5, "display": "80-85%",  "bounds": [80, 85]},
      {"value": 87.5, "display": "85-90%", "bounds": [85, 90]},
      {"value": 92.5, "display": "90-95%", "bounds": [90, 95]},
      {"value": 97.5, "display": "95-100%","bounds": [95, 100]}
    ]
  },

Comparing 8-sample datasets with and without "scale" coloring for Percent reads mapped:

Including "scale" results in expected coloring for the 4 bins. When excluding "scale", all 4 bins in the tree legend are the same color, but the tree nodes vary in color. I wonder if this is related to the fact that all 8 samples fall within one bin (95%-100%).

Comparing 8-samples versus ~1000-samples datasets for Number of drug resistance variants:

The colors of the bins for Number of drug resistance variants on the phylogenies differs for these two datasets (note that no "scale" was used for this parameter). I wonder if this is related to the fact that the 8-sample tree only has samples with up to 5 drug resistance variants (whereas the bins go up to 11+ variants), whereas the ~1000 sample tree has samples with all the different drug resistance variants categories represented.

[jameshadfield]

1. Frequencies panel shows bins based on "bounds" rather than"display"

#2015 fixed this. It's merged and will be part of the next release.

Comparing 8-sample datasets with and without "scale" coloring for Percent reads mapped

The side-by-side comparison of three datasets with slightly different configuration of the colorings dict makes clear what's happening:

In the middle and RHS screenshots, auspice is generating a scale off the values in the dataset and the minimum value is 98.37. The scale thus assigns dark blue to any value of 98 or below. The requested legend bins are anchored at [82.5, 87.5, 92.5, 97.5] which are all dark blue, and thus each legend swatch is dark blue. I think the solution is to modify the domain of the scale to encompass both the data points and all legend anchor points/bounds.

Comparing 8-samples versus ~1000-samples datasets for Number of drug resistance variants:

I presume this is the same as the above issue, namely that the scale colours are based off the data points not the defined legend anchors/bounds.