Write up a short tidy multi-assay proteomics tutorial using QFeatures

[mikelove]

This ideally would start with support for MultiAssayExperiment which would directly work for QFeatures

[lgatto]

Here's, for illustration, an example pipeline that process data in a QFeatures object:

1. Filter decoy features (i.e. those with a label of -1) in all assays.
2. Only keep features of rank 1 in all assays.
3. Keep features that have a spectrum FDR smaller than 0.05 in all assays.
4. Replace 0s by NA in all assays, to make missing values explicit.
5. Log-transform quantitative values - three new assays are created from the three initial ones and named with the “log_” prefix.
6. Aggregate each PSMs into peptide-level quantities by computing the median and ignoring missing values. The new assays names are defined as ‘peptide’ followed by the orifinal assay identifier.
7. Join the 3 peptide assays into a new assay that will contain the 33 samples. Missing values are incorporated accordingly, when a peptide is observed in part of the sets.
8. Normalise the joined peptide data using median centring.
9. Aggregate the peptide-level quantities into protein by computing the median and ignoring missing values. The new assay names will be ‘protein’.

qf |>
    filterFeatures(~ label > 0) |>           ## 1
    filterFeatures(~ rank == 1) |>           ## 2
    filterFeatures(~ spectrum_fdr < 0.05) |> ## 3
    zeroIsNA(1:3) |>                         ## 4
    logTransform(i = 1:3,                    ## 5
                 name = paste0("log_", names(qf))) |>
    aggregateFeaturesOverAssays(i = 4:6,     ## 6
                                fcol = "peptide",
                                name = sub("psm", "peptide", names(qf)),
                                fun = colMedians,
                                na.rm = TRUE) |>
    joinAssays(i = 7:9,                      ## 7
               name = "peptides") |>
    normalize(i = 10,                        ## 8
              name = "norm_peptides",
              method = "center.median") |>
    aggregateFeatures(i = "norm_peptides",   ## 9
                      name = "proteins",
                      fcol = "proteins",
                      fun = colMedians,
                      na.rm = TRUE)

From the sager vignette.

[lgatto]

And by the way, there's already a longFormat() function for MultiAssayExperiment and QFeatures objects:

> fts1
An instance of class QFeatures containing 2 assays:
 [1] assay1: SummarizedExperiment with 10 rows and 4 columns 
 [2] assay2: SummarizedExperiment with 4 rows and 4 columns 
> colData(fts1)
DataFrame with 4 rows and 2 columns
         Var1        Var2
    <numeric> <character>
S1 -1.0588267           A
S2  0.0199355           B
S3 -0.0761972           C
S4 -1.0501452           D
> longFormat(fts1, colvars = names(colData(fts1)))
DataFrame with 56 rows and 7 columns
          assay     primary  rowname  colname     value       Var1        Var2
    <character> <character> <factor> <factor> <integer>  <numeric> <character>
1        assay1          S1        a       S1         1   -1.05883           A
2        assay1          S1        b       S1         2   -1.05883           A
3        assay1          S1        c       S1         3   -1.05883           A
4        assay1          S1        d       S1         4   -1.05883           A
5        assay1          S1        e       S1         5   -1.05883           A
...         ...         ...      ...      ...       ...        ...         ...
52       assay2          S3        d       S3        12 -0.0761972           C
53       assay2          S4        a       S4        13 -1.0501452           D
54       assay2          S4        b       S4        14 -1.0501452           D
55       assay2          S4        c       S4        15 -1.0501452           D
56       assay2          S4        d       S4        16 -1.0501452           D

[mikelove]

This is great! I will change this challenge to 1) write up a short tutorial for tidyomics/tidy-proteomics showing analogy between tidySE and QFeatures, 2) allow arbitrary tidyverse verbs? e.g. mutate, group_by, summarize?