Parameters | Workflows | Normalization Method

Normalization Method
Select a normalization method. Choices are summarized in the table below. Available choices differ depending on the process.
 • No normalization.
 • Centers each column to mean zero (0) and scales each to variance one (1).
 • Centers to mean zero (0).
 • Centers to median zero (0).
 •
 • RPM (Reads Per Million) is a straightforward normalization method for Count data. It divides the raw read count by the total reads or the total mapped reads, and multiplies the result by 1,000,000.
 • Refer to the RPM Scaling process description for additional information.
 • Upper Quartile Scaling applies a scaling factor based on upper quartile to scale each column. The within column upper quartile is calculated by excluding the rows of all 0 (or missing) values. Each upper quartile is further standardized by dividing by the geometric mean among all upper quartiles across columns to be the upper quartile scaling factors.
 • Refer to the Upper Quartile Scaling process description for additional information.
 • TMM (Trimmed Mean of M component) is a scaling normalization method for RNA-Seq data. The M and A components between the targeting sample (under normalization) and the reference sample are calculated for selecting partial data to take the weighted trimmed mean of the M component as the scaling factor corresponding to the targeting sample.
 • A certain percentage of the data in the lower and higher range of the M and A components are trimmed out before taking the mean of the M component.
 • Refer to the TMM Normalization process description for additional information.
 • KMM (Kernel Density Mean of M component) is similar to TMM. The M and A components between the targeting sample (under normalization) and the reference sample are calculated for estimating the two-dimensional Kernel Density and applying the density for the weighted mean of the M component as the scaling factor corresponding to the targeting sample.
 • Refer to the KDMM Normalization process description for additional information.
 • Selects a subset as the training data based on Kernel Density.
 • Quantile Normalization is applied to the training data set, and all data except for the training data set is normalized.
 • Applies Kernel Density as a weight for Loess modeling. Click the radio button corresponding to your choice.