Impute a Peak table

imputePeakTable(peak_table, dataset, cluster_stats)

Arguments

peak_table: A matrix, with samples in rows and clusters in columns. It must have row names and column names.
dataset: The dataset object to extract samples from
cluster_stats: A data frame with the [dt|rt]_[min|max]_[ms|s] columns

Value

The imputed peak_table

Examples


# We are going to create a peak table matrix, typically resulting from [peakTable()]
# The peak table may have some missing values
# Since the missing values correspond to peaks that have not been detected
# in those particular samples, we can integrate the region where they should appear
# to get a value different than zero that reflects the noise level.
#
# Ingredients:
# - The GCIMSSample objects, so we can integrate the regions of interest (given as a GCIMSDataset)
# - The peak table matrix we want to impute
# - The definition of the regions corresponding to each cluster (cluster_stats)
#
# We will prepare here a synthetic example, check the vignette for a real use case
#
# Imagine we have information on the location of Cluster1 and Cluster2
cluster_stats <- data.frame(
  cluster = c("Cluster1", "Cluster2"),
  dt_min_ms = c(8, 10),
  dt_max_ms = c(9, 12),
  rt_min_s = c(120, 300),
  rt_max_s = c(128, 320)
)

# We have a peak table for two samples and two peaks
peak_table <- matrix(NA_real_, nrow = 2, ncol = 2)
rownames(peak_table) <- c("Sample1", "Sample2")
colnames(peak_table) <- c("Cluster1", "Cluster2")

# where we previously integrated Cluster2 in Sample 1 and Cluster1 in Sample2:
peak_table["Sample1", "Cluster2"] <- 9.5
peak_table["Sample2", "Cluster1"] <- 3.6
# The table has missing values, because some peaks were not detected.
# Maybe they are close to the noise level, or maybe they do not exist
peak_table
#>         Cluster1 Cluster2
#> Sample1       NA      9.5
#> Sample2      3.6       NA

# We will fill the missing values by integrating whatever we find
# (typically noise or small peaks) in the cluster regions of each sample. So we
# need the sample matrices.
#
# Let's build dummy Sample1 and Sample2:
## Create drift time and retention time vectors:
dt <- seq(from = 0, to = 13, by = 0.1)  # ms
rt <- seq(from = 0, to = 350, by = 1)   # s

## Create matrices with random gaussian noise.
set.seed(42)
s1_intensity <- matrix(
  rnorm(length(dt)*length(rt), sd = 0.1),
  nrow = length(dt),
  ncol = length(rt)
)
s2_intensity <- matrix(
  rnorm(length(dt)*length(rt), sd = 0.1),
  nrow = length(dt),
  ncol = length(rt)
)

# The matrix will have a pleateau in a region where the peak is supposed
# to be, so when we impute the region corresponding to Sample1-Cluster1 we see a
# higher value:
s1_intensity[dt > 8.25 & dt < 8.75, rt > 122 & rt < 126] <- 1

## Create GCIMSSample objects
s1 <- GCIMSSample(
  drift_time = dt,
  retention_time = rt,
  data = s1_intensity
)
s2 <- GCIMSSample(
  drift_time = dt,
  retention_time = rt,
  data = s2_intensity
)
## And a dataset with the samples:
dataset <- GCIMSDataset_fromList(list(Sample1 = s1, Sample2 = s2))

# Now we can impute the table
peak_table_imp <- imputePeakTable(
  peak_table = peak_table,
  dataset = dataset,
  cluster_stats = cluster_stats
)
peak_table_imp
#>         Cluster1   Cluster2
#> Sample1 1.476875  9.5000000
#> Sample2 3.600000 -0.2335684