Batch_correction

Description

Workflow position

Input files

Data Matrix file must contain the intensity values of variables.

First line must contain all the samples' names

First column must contain all the variables' ID

Sample metadata file must contain at least the three following columns:

- a batch column (default to "batch") to identify the batches of analyses

- an injection order column (default to "injectionOrder") composed of integers defining the injection order of samples

- a sample type column (default to "sampleType") indicating if a sample is a biological one ("sample"), a QC-pool ("pool") or a blank ("blank")

Default values can be changed according to your data coding using the customisation parameters in the "Sample metadata file coding parameters" section.

Notes concerning your design:

- the 3 mandatory columns must not contain NA

- your data should contain at least 3 QC-pools in each batch for intra-batch linear adjustment and 8 for lo(w)ess adjustment (minimum of 5 for all loess methods)

Parameters

Sample metadata file coding parameters

Enables to give the names of columns in the sample metadata table that contain the injection order, the batches and the sample types.

Also enables to specify the sample type coding used in the sampletype column.

Type of regression model

To choose between linear, lowess, loess, all loess pool, and all loess sample strategies

- Option 1 (linear, lowess, and loess methods): before the normalisation of each variable, some quality metrics are computed (see the "Determine Batch Correction" module); depending on the result, the variable can be normalized or not, with either the linear, lowess or loess model.

- Option 2 (all loess pool and all loess sample): each variable is normalized by using the 'loess' model;

in the case all loess pool is chosen and the number of pool observations is below 5, the linear method is used (for all variables) and a warning is generated;

if the pool intensities are not representative of the samples (which can be viewed on the figure where both trends are shown), the case all loess sample enables using the sample intensities (instead of the pool intensities) as the reference for the loess curve.

In all "option 2" cases: the median intensity of the reference observations (either 'pool' or 'sample') is used as the scaling factor after the initial intensities have been divided by the loess predictions.

Span

Smoothing parameter, advanced option for lo(w)ess and all loess methods

In case of a loess fit, the span parameter (between 0 and 1) controls the smoothing

(the higher the smoother; higher values are prefered to avoid overfitting; Cleveland et al, 1997).

Unconsistant values

available for regression model linear, lowess and loess

Controls what is done regarding negative or infinite values that can be generated during regression estimation.

Prevent it will change the normalisation term leading to an unconsistant value to prevent it:

when intra batch denominator term is below 1, it is turned to the minimum >1 one obtained in the concerned batch.

Consider it as a missing value will switch concerned intensities to NA;

this option implies that concerned ions will not be considered in PCA display.

Consider it as a null intensity will switch concerned intensities to 0.

Factor of interest

available for regression model linear, lowess and loess

Name of the factor (column header) that will be used as a categorical variable for design plots (often a biological factor ; if none, put the batch column name).

This factor does not affect correction calculation.

Level of details for plots

available for regression model linear, lowess and loess

basic: Sum of intensities + PCA + CV boxplot (before and after correction)

standard: 'basic' plots + before/after-correction plots of intensities over injection order, and design effects for each ion

complete: 'standard' plots + QC-pool regression plots per batch with samples' intensities over injection order

This factor is not used by the all loess methods where a unique figure is generated showing the sum of intensities along injection order, and the first 4 PCA scores.

Output files

Batch_correction_$method_rdata.rdata ('all_loess' methods only)

binary data

Download, open R and use the 'load' function; objects are in the 'res' list

Batch_correction_$method_graph.pdf

graphical output

For the linear and lo(w)ess methods, content depends on level of details chosen

Batch_correction_$method_variableMetadata.tabular

tsv output

Identical to the variable metadata input file, with x more columns (where x is the number of batches) in case of linear, lowess and loess methods

Batch_correction_$method_dataMatrix.tabular

tsv output (tabulated)

Same formatting as the data matrix file; contains corrected intensities

Additional information

Refer to the corresponding "W4M HowTo" page:

MS data processing - Filters and normalisation

See also the reference history:

W4M00001_Sacurine-statistics (DOI:10.15454/1.4811121736910142E12)

NEWS