What is xcms

Framework for processing and visualization of chromatographically separated and single-spectra mass spectral data.

xcms

and still growing …

RforMassSpectrometry

Outline

  • Data analysis, organization and data matrices
  • Some thoughts on validation
  • Preprocessing and analytical variability
  • MS for Dummies
  • LC-MS data handling
  • Demo & DIY
  • Peak Picking in xcms
  • Demo & DIY
  • Retention time correction and features definition
  • Demo & DIY
  • Dealing with Fragmentation esperiments
  • Demo & DIY

The data matrix

The role of Data Analysis

Statistics, Bionformatics, Machine Learning, Chemometrics, …, provide the tools to:

  • make science shared and reproducible … ;-)
  • process and organize big data into the matrix
  • identify the presence of organization in the data matrix
  • assess the confidence that our result is true “at the population level”

Examples of Organization

False Positives

  • Organization can show up only by chance
  • These results are true, but the hold only for the data we are analyzing now
  • Organization is not necessarily science
  • Variability causes this
  • We need to validate our outcomes

On Validation

  • Statistical Validation: get brand new samples and see if what we get is still there
  • Domain Validation: is what I’m getting in keeping with the domain specific body of knowledge? Could I design an experiment to check my hypothesis?

Do we always need statistics?

Data hygiene

  • Go for a scripting language and forget Excel
  • … at least use a gui pipeline or a web based solution
  • Organize data and metadata
  • Avoid as much as possible manual curation
  • Share your data, your scripts, your results
  • Go open source

Get out your data

  • Metabolomics data are always stored in “formats” which are specifically developed by instrument vendors
  • In the case of MS data several open source standards are available (cdf, mzML, mzIML, …)

Proteowizard

  • command line tool
  • gui application
  • docker with proprietary libraries

LC-MS For Dummies

Analytical Variability in LC-MS

Mass

Analytical Variability in LC-MS

Retention time

Analytical Variability in LC-MS

Intensity

Preprocessing

  • I call preprocessing all the data carpentry steps I do to go from the raw experimental data to the data matrix
  • The aim of this process is to compensate for analytical variability being able to reliably build a data matrix
  • QC samples play a big role on that because they are sensitive only to analytical variability

Uses of QCs

QCs should be representative of the chemical complexity of your samples

  • correct for retention time shifts
  • identify reliable variables:
    • variance in QC should be smaller than in samples
    • they should decrease during dilution
    • …
  • help in correcting for bath effects …

LC-MS produces 3D data

(rt,mz,I)

Things too look at

  • Extracted Ion Trace/Current (EIT/EIC)
  • Mass Spectra

Extracted ion traces

Mass Spectra

Back to Raw data

Always check your results on the raw data

  • problems in preprocessing
  • bad peaks
  • biomarkers
  • results hidden in noise

Peak Picking

Peaks and metabolites: facts

  • A metabolite produces peaks in the extracted ion traces of its associated ions
  • Different peaks in the same ion chromatograms are associated to different metabolites
  • Peaks are not metabolites
  • The same peak can slightly move across the injections

We need an automatic method to look for peaks

MatchedFilter

Anal Chem 2006 1;78(3):779-87. doi: 10.1021/ac051437y

Cent Wave

Peak Intensity: into and maxo

Things to always consider

  • Real peaks can be really badly shaped
  • You are better than an algorithm … maybe AI will do well
  • Every algorithm has parameters to tune!
  • Look to the data!
  • Know how the instrument works
  • Check what happens to metabolites you know should be there

Retention time correction and feature definition

… Just a recap

  1. We converted the data files in an open source format (here mzML)
  2. We optimized the peak picking parameters working on a representative sample (Qc)
  3. We have been running peak picking on the full set of samples
  4. We have been saving the output somewhere, just to avoid re-starting from scratch ;-)

What Next

We have to merge the lists of chromatographic peaks into a consensus list of features peaks, which will be the columns of our data matrix

  • chromatographich peaks what was detected in the individual samples (mz,rt,intensity)
  • features consensus variables which are grouping several peaks coming from the different injections (mz,rt, intensity)

Grouping - Correspondence

Grouping - Correspondence

Dynamic Time Warping

Available in xcms trough obiwarp

RT Checks

Things to always consider

  • Aligning samples and not QCs can be tricky
  • Some metabolites could not be present in pooled QC (dilutions)
  • Sometimes chromatographic peaks are missed
  • Always check the data and the known peaks!
  • Parameters are easier to tune if you know how the analytics works
  • Rt shifts should be smaller than peak width

NAssss NAssss

Even if you do everything well your final data matrix will be full of missing values:

  • errors in peak picking
  • “absence” of a metabolite in one or more samples (biology)
  • that metabolite is below the detection limit (analytics)
  • missing at random
  • missing not at random

Fragmentation Data

Annotation and MS

  • At the end of the journey we would like to work on metabolites or pathways and not on features
  • We know that annotation is the most challenging step of all the business
  • The more we know about the structure of our ions the better it is
  • Database of standards, web resources, chemoinformatics, …
  • Fragmentation patterns are extremely useful

MS/MS and DDA

DDA: Data Dependent Acquisition

Notes

  • xcms allows also to handle fragmentation experiments
  • xcms ecosystem is growing towards databases of spectra
  • As we already discussed, do not be too optimistic with complex MS experiments
  • Sometimes the ion you are interested was not fragmented (plan a specific run!)