Data normalization

To reduce technical variation between plates and studies, Olink recommends that samples are randomized, and that intensity normalization is performed before statistical analysis.

How do you normalize between plates of randomized samples?

For projects where samples are randomized, Olink recommends applying intensity normalization using the plate median as the normalization factor. The intensity normalization adjusts the data to make the median value for each assay on every plate equal across the plates. This method assumes that the actual median of each plate is very similar, which can be ensured by complete randomization of the samples. When complete randomization can be assumed, this is a robust and high performing normalization method.

For Target products, intensity normalization is performed in the following way:

  1. For each assay, the overall median value for all samples and plates is calculated.
  2. For each plate and assay, the plate-specific median value is calculated.
  3. For each assay, the plate-specific median is subtracted from every sample of the plate, this centralizes the median to 0.
  4. For each assay, the overall median value is added to every sample in the project which equals centralizing to the overall median.

For Explore products, intensity normalization is simplified and performed in the following way:

  1. For each plate and assay, the plate-specific median value is calculated.
  2. For each assay, the plate-specific median is subtracted from every sample of the plate, this centralizes the median to 0.

Figure 1. NPX across plates for an example protein before normalization

 

Figure 2. NPX across plates for a protein after intensity normalization for Target products. Note that for Explore products, the NPX values would be centered around 0.

 

How can non-randomized projects be normalized?

If the samples have not or cannot be randomized Olink does not apply any extra normalization step. However, in a project with more than one plate of samples, we recommend performing a bridge sample normalization with between 8-16 bridging samples run on all plates, to minimize technical variation. If possible, these bridge samples should be representative for the entire study, for example include all study groups, and should be matrix-matched with other samples.

NOTE: Unlike intensity normalization this type of normalization is not included in the standard analysis service.

Bridge sample normalization is performed in the following way:

  1. Choose a reference plate to normalize towards.
  2. For each assay and other project, calculate the pairwise difference for each of the overlapping samples with the reference plate.
  3. Estimate the plate- and assay-specific normalization factor by calculating the median for the pairwise differences calculated in step 2.
  4. For each assay and plate, add the plate- and assay-specific normalization factor from step 3 to each value, to normalize it to the reference plate chosen in step 1.

Figure 3. NPX across plates for an example protein including bridge samples

 

Figure 4. NPX across plates for an example protein after bridge sample normalization

Links:

Sample randomization FAQ

How can I compare results from two different studies? FAQ

How is the data pre-processed? FAQ

For more information see our white paper, Data normalization and standardization.

 

Not what you were looking for?

Ask us



2947

Biomarker assays

~881 million

Protein data points generated

1068

Publications listed on website

Contact us!