CellMix 1.6

DSA Algorithm: Digital Sorting Algorithm


  
####################################################
# Sample deconvolution analysis with DSA in CellMix
####################################################
# load benchmark data
x <- ExpressionMix('GSE19830', verbose=TRUE)



Loading dataset 'GSE19830' ... OK

dim(x)



  Features    Samples Components 
     31099         42          3 

annotation(x)



[1] "rat2302.db"

# extract mixed samples
mix <- mixedSamples(x)

# load TIGER marker list
ml <- MarkerList('TIGER')

ml




Types: bladder, blood, ..., uterus (total: 30)
Mode: numeric
setName: TIGER 
geneIds: Hs.405866, Hs.281295, ..., Hs.424414 (total: 7743)
geneIdType: Unigene (org.Hs.eg.db)
collectionType: Null 
geneValues: 99.8603, 78.4616, ..., 5.04784 
details: use 'details(object)'

names(ml)



 [1] "bladder"                   "blood"                     "bone"                      "bone_marrow"              
 [5] "brain"                     "cervix"                    "colon"                     "eye"                      
 [9] "heart"                     "kidney"                    "larynx"                    "liver"                    
[13] "lung"                      "lymph_node"                "mammary_gland"             "muscle"                   
[17] "ovary"                     "pancreas"                  "peripheral_nervous_system" "placenta"                 
[21] "prostate"                  "skin"                      "small_intestine"           "soft_tissue"              
[25] "spleen"                    "stomach"                   "testis"                    "thymus"                   
[29] "tongue"                    "uterus"                   

# select markers for the tissues present in the mixture
basisnames(x)



[1] "Brain" "Liver" "Lung" 

ml <- ml[c('brain', 'liver', 'lung')]

summary(ml)




Types: 3 ['brain', 'liver', 'lung'] 
Mode: numeric
Markers: 868 
IDtype: UNIGENE ['Hs.7124', 'Hs.12440', ..., 'Hs.228320'] 
Values: [7.26871, 7.26871, ..., 5.00013]
Source: org.Hs.eg.db 
Breakdown:
brain liver  lung 
  342   334   192 

# convert to match annotations
mlx <- convertIDs(ml, mix, verbose=TRUE)



# Converting 868 markers from Unigene (org.Hs.eg.db) to Annotation (rat2302.db) ... OK [261/868 (1:1)]
# Processing 868 markers from Unigene (org.Hs.eg.db) to Annotation (rat2302.db) ... OK [261/868 (1:1)]

summary(mlx)




Types: 3 ['brain', 'liver', 'lung'] 
Mode: numeric
Markers: 261 
IDtype: .Affymetrix ['1369882_at', '1398649_at', ..., '1370458_at'] 
Values: [7.26871, 7.26871, ..., 5.00013]
Source: rat2302.db 
Breakdown:
brain liver  lung 
  107   106    48 

# QC on markers from their expression patterns in mixed samples
profplot(mlx[,1:10], mix)


Warning message:
'x' is NULL so the result will be NULL
Warning message:
'x' is NULL so the result will be NULL
Warning message:
'x' is NULL so the result will be NULL
# filter out poor markers using SCOREM (based on linear-scale expression)
mlsc <- extractMarkers(mlx, expb(mix, 2), method='SCOREM', alpha=10^-12)

summary(mlsc)




Types: 3 ['brain', 'liver', 'lung'] 
Mode: numeric
Markers: 113 
IDtype: .Affymetrix ['1373774_at', '1378796_at', ..., '1379653_a_at'] 
Values: [0.977665444448239, 0.974762280427837, ..., 0.951324094855668]
Source: rat2302.db 
Breakdown:
brain liver  lung 
   17    81    15 

# expresison patterns are more correlated
profplot(mlsc[,1:10], mix)


Warning message:
'x' is NULL so the result will be NULL
Warning message:
'x' is NULL so the result will be NULL
Warning message:
'x' is NULL so the result will be NULL
# apply DSA using all markers
res <- ged(mix[mlsc,], mlsc, 'DSA', verbose=TRUE)



Using ged algorithm: “DSA”
     Estimating basis and mixture coefficients matrices from marker features [DSA]
     Using 113/113 markers to estimate cell proportions: 
 brain liver  lung 
    17    81    15 
      Checking data scale ...       NOTE [log]
      Converting data to linear scale ...       OK [base: 2]
      Computing proportions using DSA method ...       OK
      Estimating basis matrix from mixture coefficients [qprog]
      Not using any marker constraints
Timing:
   user  system elapsed 
  1.808   0.024   1.837 
GED final wrap up ... OK

# plot against true proportions
profplot(mix, res)