Scoring Functions
The CaDrA package currently supports four scoring functions to search for subsets of genomic features that are likely associated with a specific outcome of interest (e.g., protein expression, pathway activity, etc.)
- Kolmogorov-Smirnov Method (
ks) - Conditional Mutual Information Method (
revealer) - Wilcoxon Rank-Sum Method (
wilcox) - Custom - An User Defined Scoring Method (
custom)
Below, we run candidate_search() over the top 3 starting features using each of the four scoring functions described above.
Important Note:
- The legacy or deprecated function
topn_eval()is equivalent to the new and recommendedcandidate_search()function
Load packages
library(CaDrA)
library(pheatmap)
library(SummarizedExperiment)Load required datasets
- A
binary features matrixalso known asFeature Set(such as somatic mutations, copy number alterations, chromosomal translocations, etc.) The 1/0 row vectors indicate the presence/absence of ‘omics’ features in the samples. TheFeature Setcan be a matrix or an object of class SummarizedExperiment from SummarizedExperiment package) - A vector of continuous scores (or
Input Scores) representing a functional response of interest (such as protein expression, pathway activity, etc.)
# Load pre-computed feature set
data(sim_FS)
# Load pre-computed input scores
data(sim_Scores)Heatmap of simulated feature set
The simulated dataset, sim_FS, comprises of 1000 genomic features and 100 sample profiles. There are 10 left-skewed (i.e. True Positive or TP) and 990 uniformly-distributed (i.e. True Null or TN) features simulated in the dataset. Below is a heatmap of the first 100 features.
mat <- SummarizedExperiment::assay(sim_FS)
pheatmap::pheatmap(mat[1:100, ], color = c("white", "red"), cluster_rows = FALSE, cluster_cols = FALSE)
Search for a subset of genomic features that are likely associated with a functional response of interest using four scoring methods
1. Kolmogorov-Smirnov Scoring Method
See ?ks_rowscore for more details
ks_topn_l <- CaDrA::candidate_search(
FS = sim_FS,
input_score = sim_Scores,
method = "ks_pval", # Use Kolmogorow-Smirnow scoring function
method_alternative = "less", # Use one-sided hypothesis testing
weights = NULL, # If weights is provided, perform a weighted-KS test
search_method = "both", # Apply both forward and backward search
top_N = 3, # Evaluate top 3 starting points for the search
max_size = 10, # Allow at most 10 features in meta-feature matrix
do_plot = FALSE, # We will plot it AFTER finding the best hits
best_score_only = FALSE # Return all results from the search
)
# Now we can fetch the feature set of top N features that corresponded to the best scores over the top N search
ks_topn_best_meta <- topn_best(ks_topn_l)
# Visualize best meta-feature result
meta_plot(topn_best_list = ks_topn_best_meta)
2. Wilcoxon Rank-Sum Scoring Method
See ?wilcox_rowscore for more details
wilcox_topn_l <- CaDrA::candidate_search(
FS = sim_FS,
input_score = sim_Scores,
method = "wilcox_pval", # Use Wilcoxon Rank-Sum scoring function
method_alternative = "less", # Use one-sided hypothesis testing
search_method = "both", # Apply both forward and backward search
top_N = 3, # Evaluate top 3 starting points for the search
max_size = 10, # Allow at most 10 features in meta-feature matrix
do_plot = FALSE, # We will plot it AFTER finding the best hits
best_score_only = FALSE # Return all results from the search
)
# Now we can fetch the feature set of top N feature that corresponded to the best scores over the top N search
wilcox_topn_best_meta <- topn_best(topn_list = wilcox_topn_l)
# Visualize best meta-feature result
meta_plot(topn_best_list = wilcox_topn_best_meta)
3. Conditional Mutual Information Scoring Method
See ?revealer_rowscore for more details
revealer_topn_l <- CaDrA::candidate_search(
FS = sim_FS,
input_score = sim_Scores,
method = "revealer", # Use REVEALER's CMI scoring function
search_method = "both", # Apply both forward and backward search
top_N = 3, # Evaluate top 3 starting points for the search
max_size = 10, # Allow at most 10 features in meta-feature matrix
do_plot = FALSE, # We will plot it AFTER finding the best hits
best_score_only = FALSE # Return all results from the search
)
# Now we can fetch the ESet of top feature that corresponded to the best scores over the top N search
revealer_topn_best_meta <- topn_best(topn_list = revealer_topn_l)
# Visualize best meta-feature result
meta_plot(topn_best_list = revealer_topn_best_meta)
4. Custom - An User Defined Scoring Method
See ?custom_rowscore for more details
# A customized function using ks-test
customized_ks_rowscore <- function(FS, input_score, meta_feature=NULL, alternative="less", metric="pval"){
# Check if meta_feature is provided
if(!is.null(meta_feature)){
# Getting the position of the known meta features
locs <- match(meta_feature, row.names(FS))
# Taking the union across the known meta features
if(length(meta_feature) > 1) {
meta_vector <- as.numeric(ifelse(colSums(FS[meta_feature,]) == 0, 0, 1))
}else{
meta_vector <- as.numeric(FS[meta_feature,])
}
# Remove the meta features from the binary feature matrix
# and taking logical OR btw the remaining features with the meta vector
FS <- base::sweep(FS[-locs, , drop=FALSE], 2, meta_vector, `|`)*1
# Check if there are any features that are all 1s generated from
# taking the union between the matrix
# We cannot compute statistics for such features and thus they need
# to be filtered out
if(any(rowSums(FS) == ncol(FS))){
warning("Features with all 1s generated from taking the matrix union ",
"will be removed before progressing...\n")
FS <- FS[rowSums(FS) != ncol(FS), , drop=FALSE]
}
}
# KS is a ranked-based method
# So we need to sort input_score from highest to lowest values
input_score <- sort(input_score, decreasing=TRUE)
# Re-order the matrix based on the order of input_score
FS <- FS[, names(input_score), drop=FALSE]
# Compute the scores using the KS method
ks <- apply(FS, 1, function(r){
x = input_score[which(r==1)];
y = input_score[which(r==0)];
res <- ks.test(x, y, alternative=alternative)
return(c(res$statistic, res$p.value))
})
# Obtain score statistics
stat <- ks[1,]
# Obtain p-values and change values of 0 to the machine lowest value
# to avoid taking -log(0)
pval <- ks[2,]
pval[which(pval == 0)] <- .Machine$double.xmin
# Compute the -log(pval)
# Make sure scores has names that match the row names of FS object
pval <- -log(pval)
# Determine which metric to returned the scores
if(metric == "pval"){
scores <- pval
}else{
scores <- stat
}
names(scores) <- rownames(FS)
return(scores)
}
# Search for best features using a custom-defined function
custom_topn_l <- CaDrA::candidate_search(
FS = SummarizedExperiment::assay(sim_FS),
input_score = sim_Scores,
method = "custom", # Use custom scoring function
custom_function = customized_ks_rowscore, # Use a customized scoring function
custom_parameters = NULL, # Additional parameters to pass to custom_function
search_method = "both", # Apply both forward and backward search
top_N = 3, # Evaluate top 3 starting points for the search
max_size = 10, # Allow at most 10 features in meta-feature matrix
do_plot = FALSE, # We will plot it AFTER finding the best hits
best_score_only = FALSE # Return all results from the search
)
# Now we can fetch the feature set of top N feature that corresponded to the best scores over the top N search
custom_topn_best_meta <- topn_best(topn_list = custom_topn_l)
# Visualize best meta-feature result
meta_plot(topn_best_list = custom_topn_best_meta)
SessionInfo
sessionInfo()
R version 4.4.0 RC (2024-04-16 r86468)
Platform: x86_64-pc-linux-gnu
Running under: Ubuntu 22.04.4 LTS
Matrix products: default
BLAS: /home/biocbuild/bbs-3.20-bioc/R/lib/libRblas.so
LAPACK: /usr/lib/x86_64-linux-gnu/lapack/liblapack.so.3.10.0
locale:
[1] LC_CTYPE=en_US.UTF-8 LC_NUMERIC=C
[3] LC_TIME=en_GB LC_COLLATE=C
[5] LC_MONETARY=en_US.UTF-8 LC_MESSAGES=en_US.UTF-8
[7] LC_PAPER=en_US.UTF-8 LC_NAME=C
[9] LC_ADDRESS=C LC_TELEPHONE=C
[11] LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C
time zone: America/New_York
tzcode source: system (glibc)
attached base packages:
[1] stats4 stats graphics grDevices utils datasets methods
[8] base
other attached packages:
[1] CaDrA_1.3.0 pheatmap_1.0.12
[3] SummarizedExperiment_1.35.0 Biobase_2.65.0
[5] GenomicRanges_1.57.0 GenomeInfoDb_1.41.0
[7] IRanges_2.39.0 S4Vectors_0.43.0
[9] BiocGenerics_0.51.0 MatrixGenerics_1.17.0
[11] matrixStats_1.3.0 testthat_3.2.1.1
[13] devtools_2.4.5 usethis_2.2.3
loaded via a namespace (and not attached):
[1] bitops_1.0-7 tcltk_4.4.0 remotes_2.5.0
[4] rlang_1.1.3 magrittr_2.0.3 compiler_4.4.0
[7] vctrs_0.6.5 reshape2_1.4.4 stringr_1.5.1
[10] profvis_0.3.8 pkgconfig_2.0.3 crayon_1.5.2
[13] fastmap_1.1.1 XVector_0.45.0 ellipsis_0.3.2
[16] labeling_0.4.3 caTools_1.18.2 utf8_1.2.4
[19] promises_1.3.0 rmarkdown_2.26 sessioninfo_1.2.2
[22] UCSC.utils_1.1.0 purrr_1.0.2 xfun_0.43
[25] zlibbioc_1.51.0 cachem_1.0.8 jsonlite_1.8.8
[28] highr_0.10 later_1.3.2 DelayedArray_0.31.0
[31] parallel_4.4.0 R6_2.5.1 RColorBrewer_1.1-3
[34] bslib_0.7.0 stringi_1.8.3 pkgload_1.3.4
[37] brio_1.1.5 jquerylib_0.1.4 Rcpp_1.0.12
[40] iterators_1.0.14 knitr_1.46 R.utils_2.12.3
[43] httpuv_1.6.15 Matrix_1.7-0 R.cache_0.16.0
[46] tidyselect_1.2.1 rstudioapi_0.16.0 abind_1.4-5
[49] yaml_2.3.8 doParallel_1.0.17 gplots_3.1.3.1
[52] codetools_0.2-20 miniUI_0.1.1.1 misc3d_0.9-1
[55] pkgbuild_1.4.4 lattice_0.22-6 tibble_3.2.1
[58] plyr_1.8.9 shiny_1.8.1.1 withr_3.0.0
[61] evaluate_0.23 desc_1.4.3 urlchecker_1.0.1
[64] pillar_1.9.0 KernSmooth_2.23-22 foreach_1.5.2
[67] generics_0.1.3 rprojroot_2.0.4 ggplot2_3.5.1
[70] munsell_0.5.1 scales_1.3.0 gtools_3.9.5
[73] xtable_1.8-4 glue_1.7.0 ppcor_1.1
[76] tools_4.4.0 fs_1.6.4 grid_4.4.0
[79] colorspace_2.1-0 GenomeInfoDbData_1.2.12 cli_3.6.2
[82] fansi_1.0.6 S4Arrays_1.5.0 dplyr_1.1.4
[85] gtable_0.3.5 R.methodsS3_1.8.2 sass_0.4.9
[88] digest_0.6.35 SparseArray_1.5.0 farver_2.1.1
[91] htmlwidgets_1.6.4 memoise_2.0.1 htmltools_0.5.8.1
[94] R.oo_1.26.0 lifecycle_1.0.4 httr_1.4.7
[97] mime_0.12 MASS_7.3-60.2