differential_detection_filter: Protein differential detection filtering using a hardcoded...

In ftwkoopmans/msdap: Mass Spectrometry Downstream Analysis Pipeline

Protein differential detection filtering using a hardcoded number of peptide observations per condition

Description

Pseudocode

Number of peptides differentially detected in condition 1 as compared to condition 2:

npep_pass1 = sum(n1 - n2 >= k1_diff)

Where n1 and n2 are vectors with the number of detects per peptide in conditions 1 and 2, respectively, and k1_diff is the sample count threshold for differential detection.

Now we can test at protein-level, directionally! We're looking for proteins that are near-exclusively detected in condition 1 and their nobs ratio is in the same direction.

npep_pass1 >= npep_pass & nobs1 > nobs_ratio * nobs2

Where npep_pass and nobs_ratio are user-provided thresholds, nobs1 is the sum of n1 (see further return value specification for this function). Above rule tests enrichment in condition 1, afterwards we also test analogously with conditions 1 and 2 swapped.

real-world example data:

2 conditions, 6 replicates each:

protein_id	peptide_id	n1	n2
P09041	ALMDEVVK_2	2	6
P09041	LTLDKVDLK_2	1	5

n2-n1 diff is 4 for both peptides, nobs ratio was 3.7-fold enrichment in condition 2.

Usage

differential_detection_filter(
  dataset,
  k_diff = NA,
  frac_diff = NA,
  npep_pass = 2L,
  nobs_ratio = 3,
  int_ratio = 0,
  normalize_intensities = TRUE
)

Arguments

dataset	your dataset. At least 1 contrast should have been specified prior
k_diff	peptide-level criterium: peptide must be detected in at least k more samples in condition 1 versus condition 2, or vice versa
frac_diff	peptide-level criterium: peptide must be detected in at least x% more samples in condition 1 versus condition 2, or vice versa
npep_pass	minimum number of peptides that must pass the peptide-level differential detection criteria. Default: 2
nobs_ratio	minimum enrichment ratio between experimental conditions for the total peptide detection count per protein (see output table specification, nobs1 and log2fc_nobs2/nobs1 to which this filter is applied). Note that this value is NOT on log2 scale, i.e. set 2 for 2-fold enrichment. Default: 4
int_ratio	analogous to nobs_ratio, but for the enrichment in sum peptide intensity values. Default: 0 (disabled)
normalize_intensities	normalize the protein intensity matrix prior to computing sum intensities and intensity ratios. Default: TRUE

Value

A tibble where each row describes 1 proteingroup ("protein_id") in 1 contrast, with the following columns:

• npep_total = total number of peptides detected across any of the samples in the current contrast. Useful for post-hoc filtering, e.g. when you do not set stringent criteria for npep_pass

• npep_pass1 = number of peptides that pass the specified filtering rules in condition 1 of the current contrast

• npep_pass2 = analogous to npep_pass1, but for condition 2

• nobs1 = sum of peptide detections across all samples in condition 1 (i.e. each detected peptide is counted once per sample, this is the total sum across peptides*samples for respective protein_id and condition). Useful for post-hoc filtering, e.g. when you do not set stringent criteria for npep_pass

• nobs2 = analogous to nobs1, but for condition 2

• fracobs1 = percentage of all possible detections made within condition 1 = number of observed datapoints / (#peptide * #sample)

• fracobs2 = analogous to fracobs1, but for condition 2

• log2fc_nobs2/nobs1 = log2 foldchange of observation counts. Positive values are enriched in condition 2. Proteins exclusive to condition 2 have value Inf and exclusive to condition 1 have value -Inf

• int1 = sum peptide intensity across all samples in condition 1

• int2 = sum peptide intensity across all samples in condition 2

• log2fc_int2/int1 = log2 foldchange of protein intensities, analogous to log2fc_nobs2/nobs1

• pass = protein matches input filtering

Examples

## Not run: 
## example 1:
# default / stringent: find proteins that have at least 2 peptides
# detected in 66% more samples in either condition (with a minimum of 3 samples)
# AND the overall detection rate is larger than 3-fold
x = differential_detection_filter(
  dataset, k_diff = 3, frac_diff = 0.66, npep_pass = 2, nobs_ratio = 3
)

# code snippet to create a pretty-print table with resulting proteins
y = x %>%
  # only retain proteins that match input criteria
  filter(pass) %>%
  # add protein metadata and rearrange columns
  left_join(dataset$proteins %>%
    select(protein_id, fasta_headers, gene_symbols_or_id),
    by = "protein_id") %>%
  select(contrast, protein_id, fasta_headers, gene_symbols_or_id,
         tidyselect::everything()) %>%
  # for prettyprint, trim the contrast names
  mutate(contrast = gsub(" *#.*", "", sub("^contrast: ", "", contrast))) %>%
  # sort data by column, then by ratio
  arrange(contrast, `log2fc_nobs2/nobs1`)
print(y)

## example 2:
# a more lenient filter: apply criteria to only 1 (or 0) peptides but
# rely mostly on the nobs_ratio criterium and additionally
# add post-hoc filtering on the total number of peptides per protein
# and require either protein to have an overall 50% detection rate
# (across peptides and samples)
x = differential_detection_filter(
  dataset, nobs_ratio = 3, npep_pass = 1, k_diff = 3, frac_diff = 0.66
) %>%
  mutate(pass = pass & npep_total > 1 & pmax(fracobs1, fracobs1) >= 0.5)

## End(Not run)

ftwkoopmans/msdap documentation built on March 5, 2025, 12:15 a.m.