Mass spectrometry-based proteomics

class: center, middle, inverse, title-slide

.title[
# Mass spectrometry-based proteomics
]
.subtitle[
## Statistical Data Analysis for Genome-Scale Biology
]
.author[
### Laurent Gatto - @lgatto@fosstodon.org
]
.date[
### 27 June 2024, Brixen
]

---

## On the menu

.left-col-50[

1. The **R for Mass Spectrometry** initiative
2. Mass spectrometry-based proteomics
3. MS-based single-cell proteomics
4. Lab intro

]

.right-col-50[

### Slides available (CC-BY-SA) at [https://bit.ly/CSAMA2024](https://bit.ly/CSAMA2024)

]

???

# Mention excellent menu

---
class: middle, center, inverse

# R for Mass Spectrometry

.left-col-75[

## R software for the analysis and interpretation of high throughput mass spectrometry assays

]

.right-col-25[

<img src="https://github.com/rformassspectrometry/stickers/raw/master/sticker/RforMassSpectrometry.png" width="80%" style="display: block; margin: auto;" />
]

### www.rformassspectrometry.org

---
background-image: url("./figs/rforms.png")
background-size: contain

???

- **Home**: The aim of the RforMassSpectrometry initiative is to
  provide efficient, thoroughly documented, tested and flexible R
  software for the analysis and interpretation of high throughput mass
  spectrometry assays, including proteomics and metabolomics
  experiments. The project formalises the longtime collaborative
  development efforts of its core members under the
  RforMassSpectrometry organisation to facilitate dissemination and
  accessibility of their work.
- All packages get submitted to **Bioconductor**.
- **Packages** - and see next slide
- **Contributors** -

---
background-image: url("./figs/r4mspkgs.png")
background-size: contain

???
- Go back to homepage to mention contributors

---
class: middle, center, inverse

# How does mass spectrometry work?

---
class: middle

## Motivation

???

Coming back to a recurring theme in this course, what's the
correlation between RNA and proteins? I would like to step back a bit,
and invite you to think about the data underlying these correlations.

From a 2011 study that compared the expression profiles of 3 cell
lines using RNA-Seq, MS-based proteomics and immunofluorescence
(protein-specific antibodies).

They observed an overall Spearman correlation of 0.63.

**In what ways to these summaries differ?**

Using a common gene-centric identifier, but

- What do we have along the rows, what are our features? Transcripts
  on the left. Protein groups on the right.
- How are these intensities produced?

## Take-home message

These data tables are less similar than they appear.

---
class: middle

### Mass spectrometry-based proteomics

---
### How does MS work?

1. Digestion of proteins into peptides - as will become clear later,
   the features we measure in shotgun (or bottom-up) *proteomics* are
   peptides, **not** proteins.

2. On-line liquid chromatography (LC-MS)

3. Mass spectrometry (MS) is a technology that **separates** charged
   molecules (ions, peptides) based on their mass to charge ratio
   (M/Z).

---
### Chromatography

MS is generally coupled to chromatography (liquid LC, but can also be
gas-based GC). The time an analytes takes to elute from the
chromatography column is the **retention time**.

???

- This is an acquisition. There can be one per sample (with muliple
  fractions), of samples can be combined/multiplexed and acquired
  together.

---

A mass spectrometer is composed of three parts

.left-col-30[

1. The **source**, that ionises the molecules: examples are Matrix-assisted
   laser desorption/ionisation (MALDI) or electrospray ionisation
   (ESI).
2. The **analyser**, that separates the ions: Time of flight (TOF) or Orbitrap.
3. The **detector** that quantifies the ions.

]

.right-col-70[

<img src="./figs/SchematicMS2.png" width="85%" style="display: block; margin: auto;" />
]

???

- Important point here: if a peptides does not ionise (well), it won't
  "fly" in analyser, and thus won't be detected. -> **missing values**

---
class: middle center

![Separation and detection of ions in a mass spectrometer.](./figs/mstut.gif)

---

A mass spectrometer is composed of three parts

.left-col-30[

]

.right-col-70[

<img src="./figs/SchematicMS2.png" width="85%" style="display: block; margin: auto;" />
]

In **data dependent acquisition** (DDA), *individual* ions go through
that cylce at least twice (MS2, tandem MS, or MSMS). Before the second
cycle, individual *precursor* ions a selected and broken into
*fragment* ions (next slides).

---
class: middle

.left-col-75[
<img src="./figs/MS1-MS2-spectra.png" width="90%" style="display: block; margin: auto;" />
]

.right-col-25[

Data dependent acquisition (DDA):

- parent ions in the MS1 spectrum (left)
- selected fragment ions MS2 spectra (right).

]

???

Highlight
- Some peptides might not ionise (well) -> no ions -> **missing data**
- Low abundanec (too low, below limit of detection) -> **missing data**
- Semi-stochastic nature of MS -> **missing data**
- Computational "drop-outs" (see later) -> **missing data**
- Data dependent acquisition (DDA)

Please keep this MS space and the semi-stochastic nature of MS
acquisition in mind , as I will come back to it later.

---
class: middle, center, inverse

# Raw MS data

---
class: middle center

.left-col-50[
![MS1 spectra over retention time.](./figs/F02-3D-MS1-scans-400-1200-lattice.png)
]

.right-col-50[
![MS2 spectra interleaved between two MS1 spectra.](./figs/F02-3D-MS1-MS2-scans-100-1200-lattice.png)
]

???

- Please keep this MS space and the semi-stochastic nature of MS
  acquisition in mind , as I will come back to it later.

---
class: middle

## Spectra objects

<img src="https://rformassspectrometry.github.io/book/img/raw.png" width="95%" style="display: block; margin: auto;" />
---
class: middle

.left-col-75[

``` r
> library(Spectra)
> basename(fls <- dir(system.file("sciex", package = "msdata"),
 full.names = TRUE))
[1] "20171016_POOL_POS_1_105-134.mzML"
[2] "20171016_POOL_POS_3_105-134.mzML"
> Spectra(fls)
MSn data (Spectra) with 1862 spectra in a MsBackendMzR backend:
 msLevel rtime scanIndex
 <integer> <numeric> <integer>
1 1 0.280 1
2 1 0.559 2
3 1 0.838 3
4 1 1.117 4
5 1 1.396 5
... ... ... ...
1858 1 258.636 927
1859 1 258.915 928
1860 1 259.194 929
1861 1 259.473 930
1862 1 259.752 931
 ... 33 more variables/columns.

file(s):
20171016_POOL_POS_1_105-134.mzML
20171016_POOL_POS_3_105-134.mzML
```
]

.right-col-25[

<img src="https://github.com/rformassspectrometry/Spectra/raw/master/logo.png" width="90%" style="display: block; margin: auto;" />
]

---
class: middle, center, inverse

# Peptide identification

???

Using MS2 scans

---
###  Identification: fragment ions

<img src="./figs/frag.png" width="95%" style="display: block; margin: auto;" />
???

- Nitrogen

---
###  Identification: Peptide-spectrum matching (PSM)

Matching **expected** and *observed* spectra:

``` r
> PSMatch::calculateFragments("SIGFEGDSIGR")
           mz  ion type pos z        seq
1    88.03931   b1    b   1 1          S
2   201.12337   b2    b   2 1         SI
3   258.14483   b3    b   3 1        SIG
4   405.21324   b4    b   4 1       SIGF
5   534.25583   b5    b   5 1      SIGFE
6   591.27729   b6    b   6 1     SIGFEG
7   706.30423   b7    b   7 1    SIGFEGD
8   793.33626   b8    b   8 1   SIGFEGDS
9   906.42032   b9    b   9 1  SIGFEGDSI
10  963.44178  b10    b  10 1 SIGFEGDSIG
11  175.11895   y1    y   1 1          R
12  232.14041   y2    y   2 1         GR
13  345.22447   y3    y   3 1        IGR
14  432.25650   y4    y   4 1       SIGR
15  547.28344   y5    y   5 1      DSIGR
16  604.30490   y6    y   6 1     GDSIGR
 [ reached getOption("max.print") -- omitted 16 rows ]
```

---
class: middle

###  Identification: Peptide-spectrum matching (PSM)

Matching *expected* and **observed** spectra:

???

Performed by **search engines** such as Mascot, MSGF+, Andromeda, ...

---
### Identification: database

![Uniprot human proteome](./figs/uniprot1.png)

???

## Where do the peptides come from?

### [Human proteome](https://www.uniprot.org/help/human_proteome)

In 2008, a draft of the complete human proteome was released from
UniProtKB/Swiss-Prot: the approximately 20,000 putative human
protein-coding genes were represented by one UniProtKB/Swiss-Prot
entry, tagged with the keyword 'Complete proteome' and later linked to
proteome identifier UP000005640. This UniProtKB/Swiss-Prot H. sapiens
proteome (manually reviewed) can be considered as complete in the
sense that it contains one representative (**canonical**) sequence for
each currently known human gene. Close to 40% of these 20,000 entries
contain manually annotated alternative isoforms representing over
22,000 additional sequences

## [What is the canonical sequence?](https://www.uniprot.org/help/canonical_and_isoforms)

To reduce redundancy, the UniProtKB/Swiss-Prot policy is to describe
all the protein products encoded by one gene in a given species in a
single entry. We choose for each entry a canonical sequence based on
at least one of the following criteria:

1. It is the most prevalent.
2. It is the most similar to orthologous sequences found in other species.
3. By virtue of its length or amino acid composition, it allows the
   clearest description of domains, isoforms, polymorphisms,
   post-translational modifications, etc.
4. In the absence of any information, we choose the longest sequence.

## Are all isoforms described in one UniProtKB/Swiss-Prot entry?

Whenever possible, all the protein products encoded by one gene in a
given species are described in a single UniProtKB/Swiss-Prot entry,
including isoforms generated by alternative splicing, alternative
promoter usage, and alternative translation initiation (*). However,
some alternative splicing isoforms derived from the same gene share
only a few exons, if any at all, the same for some 'trans-splicing'
events. In these cases, the divergence is obviously too important to
merge all protein sequences into a single entry and the isoforms have
to be described in separate 'external' entries.

## UniProt [downloads](https://www.uniprot.org/downloads)

---
class: middle

### Identification: decoy to control false identification

From Käll *et al.* [Posterior Error Probabilities and False Discovery
Rates: Two Sides of the Same Coin](https://pubs.acs.org/doi/abs/10.1021/pr700739d).

???

- (global) FDR = B/(A+B)
- (local) fdr = PEP = b/(a+b)

This is the same as Wolfgang's slide: Basic problem: binary decision

---
###  Identification: Protein inference

.left-col-25[
- Keep only reliable peptides
- From these peptides, infer proteins.
- If proteins can't be resolved due to shared peptides, merge them
  into **protein groups** of indistinguishable or non-differentiable
  proteins.

]

.right-col-75[
<img src="./figs/nbt0710-647-F2.gif" width="100%" style="display: block; margin: auto;" />
From [Qeli and Ahrens
(2010)](http://www.ncbi.nlm.nih.gov/pubmed/20622826).

]

---
class: middle

``` r
> library(PSMatch)
> basename(idf <- msdata::ident(full.names = TRUE))
[1] "TMT_Erwinia_1uLSike_Top10HCD_isol2_45stepped_60min_01-20141210.mzid"
> PSM(idf)
PSM with 5802 rows and 35 columns.
names(35): sequence spectrumID ... subReplacementResidue subLocation
```

---
class: middle, center, inverse

# Quantitation

---

### Quantitation

|    |Label-free |Labelled   |
|:---|:----------|:----------|
|MS1 |XIC        |SILAC, 15N |
|MS2 |Counting   |iTRAQ, TMT |

- Quantitation data at the PSM/precursor or peptide level.
- Data need to be *aggregated* into protein level data.

???

- Decision on quant method will depend on what your core facility is
  expert in and the experimental design.

---
class: middle

### Aggregation

---
class: middle

### Aggregation

---
class: middle

### Aggregation

---
class: middle

## QFeatures objects

.left-col-75[

``` r
> library(QFeatures)
> data(hlpsms)
> hl <- readQFeatures(hlpsms, ecol = 1:10, name = "psms")
> An instance of class QFeatures containing 1 assays:
 [1] psms: SummarizedExperiment with 3010 rows and 10 columns
> hl <- aggregateFeatures(hl, "psms", "Sequence",
 name = "peptides", fun = colMeans) |>
 aggregateFeatures("peptides", "ProteinGroupAccessions",
 name = "proteins", fun = colMeans)
> hl
An instance of class QFeatures containing 3 assays:
 [1] psms: SummarizedExperiment with 3010 rows and 10 columns
 [2] peptides: SummarizedExperiment with 2923 rows and 10 columns
 [3] proteins: SummarizedExperiment with 1596 rows and 10 columns
```

]

.right-col-25[

<img src="https://raw.githubusercontent.com/rformassspectrometry/stickers/master/QFeatures/QFeatures.png" width="80%" style="display: block; margin: auto 0 auto auto;" />
]

???

- Using `SummarizedExperiment` class.

---
class: middle

### MS workflow (with packages)

---
class: middle, center, inverse

.left-col-75[

# Mass spectrometry-based
# single-cell proteomics

]

.right-col-25[

<img src="figs/scp.png" width="80%" style="display: block; margin: auto;" />
]

---
class: middle

.left-col-50[

- **August 2019**: in a Nature Methods Technology Feature
  ([10.1038/s41592-019-0540-6](https://www.nature.com/articles/s41592-019-0540-6)),
  Vivien Marx *dreamt of single-cell proteomics*.

- **March 2023**: Nature Methods published a special issue with a
  *Focus on single-cell proteomics*
  ([www.nature.com/collections/bdfhafhdeb](www.nature.com/collections/bdfhafhdeb)).

]

.right-col-50[

]

???

- Become relevant in recent years, say 2019

- Reminder: no PCR for proteins/peptides

- Possible through better sample preparation, reduction of loss of
  material, miniaturisation, automation, better MS, greater
  sensitivity, DDA and DIA, LFQ and labelling, ...

---

## Single-cell omics

| | FC | scRNA-Seq | SCP |
|:---------|:---------------------|:----------------------|-------------------|
| features | 10 | `$10^{4}$` | `$10^{3}$` |
| cells | `$10^{6}$` | `$10^{4}$` | `$10^{3}$` |
| samples | 10 - 100 | 1 - 10 | 1 ... |
| | sample throughput | feature throughput | functional PTM |

???

- RNA = intention vs. Protein = action
- Inference of direct regulatory interactions with minimal assumptions
  (Slavov:2022).

---

## SCP

### Data processing/analysis

- *same* as previously

### With substantial exacerbation of

- Batch effects (1 or 16 cells par MS acquisition) ([Vanderaa and
  Gatto (2021)](http://dx.doi.org/10.1080/14789450.2021.1988571))
- Missing values ([Vanderaa and Gatto
  (2023)](https://pubs.acs.org/doi/10.1021/acs.jproteome.3c00227))

---
class: middle

Data from [Specht et al. (2021)](https://genomebiology.biomedcentral.com/articles/10.1186/s13059-021-02267-5).

???

- Seminal dataset published by (Specht et al. (2021))
- 1096 macrophages, 394 monocytes (after QC)
- 9354 peptides, 3042 proteins
- Pre-print, data and code available since 2019
---
class: middle

Data from [Specht et al. (2021)](https://genomebiology.biomedcentral.com/articles/10.1186/s13059-021-02267-5).

---
class: center, middle, inverse

# What do others do?

---
class: center

Systematic review of SCP data analysis ([Vanderaa and Gatto
(2024)](https://currentprotocols.onlinelibrary.wiley.com/doi/10.1002/cpz1.658)).

???

Diversity wouldn't be a bad thing, but, as you might expect here,
different analysis workflows lead to different and incompatible
results.

---
class: center, middle, inverse

# A principled approach

---

### Software and data

- `scp`: https://bioconductor.org/packages/scp (uses
  [QFeatures](https://bioconductor.org/packages/QFeatures) and
  [SingleCellExperiment](https://bioconductor.org/packages/SingleCellExperiment))
- `scpdata`: https://bioconductor.org/packages/scpdata

### Minimal processing

- Remove low quality precursors and cells
- Aggregate from precursors into peptides
- log-transform
- (Remove features with too many NAs)
- No imputation

### ANOVA–simultaneous component analysis

- Linear model, including biological and technical effects
- Use effect matrices for downstream applications (PCA, differential
  abundance analysis, ...).

---

---

---
class: middle

$$ y = {\color{Blue}{MS~acquisition}} + {\color{Blue}{TMT~channel}} + {\color{Orange}{Cell~type}} + {\color{red} \epsilon} $$

Data from [Leduc et al
(2022)](https://genomebiology.biomedcentral.com/articles/10.1186/s13059-022-02817-5). **scplainer**:
using linear models to understand mass spectrometry-based single-cell
proteomics data ([Vanderaa and Gatto
(2024)](https://www.biorxiv.org/content/10.1101/2023.12.14.571792v2)).

???

- Different **effect matrices** from the coefficients.
- Quantify the contribution of these different **known** (technical
  and biological) and **unknown** effects.
- Can also be used to **integrate data from different experiments** such as DIA and DDA.

---

Integration of the plexDIA data sets by Leduc et al. and Derks et
al. ([Vanderaa and Gatto
(2024)](https://www.biorxiv.org/content/10.1101/2023.12.14.571792v2)).

---

**scplainer** method (**scp** package), described in [Vanderaa and
Gatto
(2024)](https://www.biorxiv.org/content/10.1101/2023.12.14.571792v2).

---
class: middle, center, inverse

# Proteomics practicals

---
class: middle

## R for Mass Spectrometry book

#### https://rformassspectrometry.github.io/book/

???

- Chapters on raw data, identification, quantitative data.
- Last chapter has links to specific workshops on proteomics and
  metabolomics.
- If you are interested in a package in particular (Spectra, PSMatch,
  QFeatures, ...), have a look at the vignettes.

---
class: middle

.left-col-50[

## EuroBioc2024
## Oxford, 4 - 6 September
## [eurobioc2024.bioconductor.org](https://eurobioc2024.bioconductor.org/)

]

.right-col-50[

]

---
class: center, middle

# Questions?