title: “adverSCarial, generate and analyze the vulnerability of scRNA-seq classifiers to adversarial attacks” shorttitle: “adverSCarial” author: Ghislain FIEVET ghislain.fievet@gmail.com package: adverSCarial abstract: > adverSCarial is an R Package designed for generating and analyzing the vulnerability of scRNA-seq classifiers to adversarial attacks. The package is versatile and provides a format for integrating any type of classifier. It offers functions for studying and generating two types of attacks, min change attack and max change attack. The single gene attack involves making a small modification to the input to alter the classification. The max change attack involves making a large modification to the input without changing its classification. The package provides a comprehensive solution for evaluating the robustness of scRNA-seq classifiers against adversarial attacks. vignette: > %\VignetteIndexEntry{Vign02_overView_analysis} %\VignetteEngine{knitr::knitr} %\VignetteEncoding{UTF-8}
library(adverSCarial)
library(LoomExperiment)
library(DelayedArray)
pbmcPath <- system.file("extdata", "pbmc_short.loom", package="adverSCarial")
lfile <- import(pbmcPath, type="SingleCellLoomExperiment")
matPbmc <- counts(lfile)
matPbmc[1:5,1:5]
## <5 x 5> DelayedMatrix object of type "integer":
## NEK11 IL5RA OR4C5 KHDRBS3 OPA3
## [1,] 0 0 0 0 0
## [2,] 0 0 0 0 0
## [3,] 0 0 0 0 0
## [4,] 0 0 0 0 0
## [5,] 0 0 0 0 0
cellTypes <- rowData(lfile)$cell_type
head(cellTypes)
## [1] "Naive CD4 T" "Naive CD4 T" "FCGR3A+ Mono" "Naive CD4 T" "B"
## [6] "Naive CD4 T"
singleGeneOverview and maxChangeOverviewThe singleGeneOverview and maxChangeOverview functions are designed to provide insight into of the min and max change adversarial attacks on each cell type, on various gene modifications.
Before generating an attack it is judicious to choose the cell type to attack, and the modification susceptible to lead to a successful attack. Both functions run attack approximations, faster than the original, by studying splices of 100 genes.
Sometimes we want to compare two classifiers and see which one is more vulnerable to adversarial attacks.
Here we define the modifications to analyse, the predefined perc1, and a custom function returning high outliers called modifOutlier.
modifOutlier <- function(x, y){
return (max(x)*1000)
}
modifications <- list(c("perc1"), c("full_row_fct", modifOutlier))
We run the singleGeneOverview, this gives us a general idea of which cell types are more vulnerable to single gene attacks.
min_change_overview <- singleGeneOverview(matPbmc, cellTypes, MClassifier,
modifications= modifications, maxSplitSize = 20, firstDichot = 5)
min_change_overview
## DataFrame with 10 rows and 2 columns
## perc1
## <numeric>
## Naive CD4 T 1
## FCGR3A+ Mono 1
## B 1
## Memory CD4 T 1
## CD14+ Mono 1
## CD8 T 1
## UNDETERMINED 0
## NK 1
## Platelet 1
## DC 1
## full_row_fct_function..x..y.........return.max.x....1000...
## <numeric>
## Naive CD4 T 5
## FCGR3A+ Mono 5
## B 1
## Memory CD4 T 0
## CD14+ Mono 2
## CD8 T 4
## UNDETERMINED 8
## NK 3
## Platelet 7
## DC 6
And the maxChangeOverview, giving us a general idea of which cell types are more vulnerable to max change attacks.
max_change_overview <- maxChangeOverview(matPbmc, cellTypes, MClassifier,
modifications= modifications, maxSplitSize = 20)
max_change_overview
## DataFrame with 10 rows and 2 columns
## perc1
## <numeric>
## Naive CD4 T 1
## FCGR3A+ Mono 1
## B 1
## Memory CD4 T 1
## CD14+ Mono 1
## CD8 T 1
## UNDETERMINED 1
## NK 1
## Platelet 1
## DC 1
## full_row_fct_function..x..y.........return.max.x....1000...
## <numeric>
## Naive CD4 T 135
## FCGR3A+ Mono 123
## B 187
## Memory CD4 T 200
## CD14+ Mono 174
## CD8 T 148
## UNDETERMINED 86
## NK 161
## Platelet 98
## DC 110
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] DelayedArray_0.31.0 SparseArray_1.5.0
## [3] S4Arrays_1.5.0 abind_1.4-5
## [5] Matrix_1.7-0 LoomExperiment_1.23.0
## [7] BiocIO_1.15.0 rhdf5_2.49.0
## [9] SingleCellExperiment_1.27.0 SummarizedExperiment_1.35.0
## [11] Biobase_2.65.0 GenomicRanges_1.57.0
## [13] GenomeInfoDb_1.41.0 IRanges_2.39.0
## [15] MatrixGenerics_1.17.0 matrixStats_1.3.0
## [17] S4Vectors_0.43.0 BiocGenerics_0.51.0
## [19] adverSCarial_1.3.0
##
## loaded via a namespace (and not attached):
## [1] jsonlite_1.8.8 compiler_4.4.0 crayon_1.5.2
## [4] stringr_1.5.1 rhdf5filters_1.17.0 yaml_2.3.8
## [7] lattice_0.22-6 R6_2.5.1 XVector_0.45.0
## [10] commonmark_1.9.1 knitr_1.46 GenomeInfoDbData_1.2.12
## [13] rlang_1.1.3 stringi_1.8.3 HDF5Array_1.33.0
## [16] xfun_0.43 cli_3.6.2 magrittr_2.0.3
## [19] Rhdf5lib_1.27.0 zlibbioc_1.51.0 grid_4.4.0
## [22] markdown_1.12 lifecycle_1.0.4 glue_1.7.0
## [25] evaluate_0.23 httr_1.4.7 tools_4.4.0
## [28] UCSC.utils_1.1.0