Statistical analysis · step 1

Analysis table#

Analysis table
Compare groups
Correlate

Use this when your data has gone through preprocessing and you want to analyse it: the analysis table has one row per sample and one column per assay, with your sample metadata in front. Every statistics function works on it.

import signatomics.olink as so

data = so.prepare([so.example_file(1), so.example_file(2)], so.example_samples(), on="sample_id")
table = so.to_wide(data)

On the synthetic example files

Analysis table
  averaged: 12 samples measured more than once (bridge samples or repeats) get the average of their NPX values
  result: 59 samples x 92 assays, plus 4 columns of sample metadata (sample_id, group, age, sex)

The long table becomes wide: each sample is now one row, so every sample counts once in the statistics. The 12 bridge samples were measured on both plates; after bridging, their two values are averaged. Controls are left out.

       sample_id    group  age     sex   IL8  VEGFA  CD8A  ...
sample
P01          P01  control   73  female  4.28   7.47  7.16
P02          P02     case   60  female  6.75   8.00  8.20
P03          P03  control   63  female  6.13   7.42  7.73

Remove what quality control flagged#

Quality control found two broken samples, P07 and P33: both read about 5 NPX too low on every assay. prepare leaves that decision to you. Before the statistics, remove them:

table = table.drop(["P07", "P33"])

Compare groups shows what happens when they stay: both are controls, so almost every assay would look higher in cases.

Can I keep P33's good measurement?

Yes. P33 is a bridge sample: only its measurement on Plate 2 is broken. Remove just that one before bridging, and its Plate 1 value stays:

clean = so.apply_qc(data, drop_samples={"P33": "Plate 2"})

Can I see which values are below the LOD, per sample?

below = so.to_wide(data, values="below_lod")    # True where the value is below the LOD

Give it to compare_groups(..., below_lod=below) to see the share below the LOD per group.

Check yourself: why does the analysis table come last, after bridging?

Each bridge sample has two measurements, one per plate. Only after bridging are they on the same scale, so their average is meaningful.

Every option: to_wide in Functions.

← Sample metadata Compare groups →