Within and between batch correction of LC-MS metabolomics data

This is package contains functions within three areas of batch correction. These algorithms were originally developed to increase quality and information content in data from LC-MS metabolomics. However, the algorithms should be applicable to other data structures/origins, where within and between batch irregularities occur.

The three areas indicated are:

• alignBatches(): Align features systematically misaligned between batches
• correctDrift(): Perform within-batch intensity drift correction
• normalizeBatches(): Perform between-batch normalization

Batch alignment is achieved based on three concepts:

• Aggregation of feature presence/missingness on batch level.
• Identifying features with missingness within "the box", i.e. sufficiently similar in retention time and m/z.
• Ensuring orthogonal batch presence among feature alignment candidates.

Drift correction is achieved based on:

• Clustering is performed on features in observation space (as opposed to the normally used observations in feature space)
• Clustering provides a tradeoff between
  • modelling detail (multiple drift patterns within data set)
  • power per drift pattern
• Unbiased clustering is achieved using the Bayesian mclust R package

Batch normalisation is achieved based on:

• QC/Reference (standard normalisation) or population (median normalisation)
• The choice between the two is based on a quality heuristic determining whether the QC/Ref samples are suitable for normalization. Otherwise population normalization is performed instead.

The development and inner workings of these algorithms are reported in: Brunius C, Shi L, Landberg R, 2016. Large-scale untargeted LC-MS metabolomics data correction using between-batch feature alignment and cluster-based within-batch signal intensity drift correction. Metabolomics 12:173, doi: 10.1007/s11306-016-1124-4