I'm using DiffBind for ATAC-seq DA analysis. The MA plot and volcano look nothing like I expect. I do not expect a large difference between my experimental (FRDA) and control group. But as you can see in the figures, below, there is a heavy skew for regions down regulated compared to control.

I tried loess curve offset normalization, since the curve doesn't look centered at all counts_NP.loess <- dba.normalize(counts_NP, method = DBA_ALL_METHODS ,offsets = TRUE). After adjusting the offset, the plots look even worse!

I'm fairly new to both ATAC and DiffBind, so any help and tips would be greatly appreciated!

> sessionInfo()
R version 4.2.1 (2022-06-23)
Platform: x86_64-apple-darwin17.0 (64-bit)
Running under: macOS Ventura 13.4

Matrix products: default
LAPACK: /Library/Frameworks/R.framework/Versions/4.2/Resources/lib/libRlapack.dylib

locale:
[1] en_US.UTF-8/en_US.UTF-8/en_US.UTF-8/C/en_US.UTF-8/en_US.UTF-8

attached base packages:
[1] stats4    stats     graphics  grDevices utils     datasets  methods   base     

other attached packages:
 [1] csaw_1.30.1                 GreyListChIP_1.28.1         DiffBind_3.6.5              SummarizedExperiment_1.28.0
 [5] Biobase_2.58.0              MatrixGenerics_1.10.0       matrixStats_0.63.0          GenomicRanges_1.50.1       
 [9] GenomeInfoDb_1.34.4         IRanges_2.32.0              S4Vectors_0.36.1            BiocGenerics_0.44.0        

loaded via a namespace (and not attached):
  [1] utf8_1.2.3               spatstat.explore_3.2-1   reticulate_1.28          tidyselect_1.2.0        
  [5] RSQLite_2.3.1            AnnotationDbi_1.60.0     htmlwidgets_1.6.2        grid_4.2.1              
  [9] BiocParallel_1.32.4      Rtsne_0.16               munsell_0.5.0            codetools_0.2-19        
 [13] ica_1.0-3                interp_1.1-4             systemPipeR_2.2.2        future_1.32.0           
 [17] miniUI_0.1.1.1           withr_2.5.0              spatstat.random_3.1-5    colorspace_2.1-0        
 [21] progressr_0.13.0         knitr_1.43               rstudioapi_0.14          Seurat_4.3.0            
 [25] ROCR_1.0-11              tensor_1.5               listenv_0.9.0            labeling_0.4.2          
 [29] bbmle_1.0.25             GenomeInfoDbData_1.2.9   mixsqp_0.3-48            hwriter_1.3.2.1         
 [33] polyclip_1.10-4          farver_2.1.1             bit64_4.0.5              coda_0.19-4             
 [37] parallelly_1.36.0        vctrs_0.6.2              generics_0.1.3           xfun_0.39               
 [41] R6_2.5.1                 apeglm_1.18.0            invgamma_1.1             locfit_1.5-9.7          
 [45] cachem_1.0.8             bitops_1.0-7             spatstat.utils_3.0-3     DelayedArray_0.24.0     
 [49] promises_1.2.0.1         BiocIO_1.8.0             scales_1.2.1             gtable_0.3.3            
 [53] globals_0.16.2           goftest_1.2-3            rlang_1.1.1              splines_4.2.1           
 [57] rtracklayer_1.58.0       lazyeval_0.2.2           spatstat.geom_3.2-1      yaml_2.3.7              
 [61] reshape2_1.4.4           abind_1.4-5              httpuv_1.6.11            tools_4.2.1             
 [65] ggplot2_3.4.2            ellipsis_0.3.2           gplots_3.1.3             RColorBrewer_1.1-3      
 [69] ggridges_0.5.4           Rcpp_1.0.10              plyr_1.8.8               zlibbioc_1.44.0         
 [73] purrr_1.0.1              RCurl_1.98-1.12          deldir_1.0-9             pbapply_1.7-0           
 [77] ashr_2.2-54              cowplot_1.1.1            zoo_1.8-12               SeuratObject_4.1.3      
 [81] ggrepel_0.9.3            cluster_2.1.4            magrittr_2.0.3           data.table_1.14.8       
 [85] scattermore_1.1          lmtest_0.9-40            RANN_2.6.1               truncnorm_1.0-9         
 [89] mvtnorm_1.1-3            SQUAREM_2021.1           amap_0.8-19              fitdistrplus_1.1-11     
 [93] patchwork_1.1.2          mime_0.12                evaluate_0.21            xtable_1.8-4            
 [97] XML_3.99-0.14            emdbook_1.3.12           jpeg_0.1-10              gridExtra_2.3           
[101] compiler_4.2.1           bdsmatrix_1.3-6          tibble_3.2.1             KernSmooth_2.23-21      
[105] crayon_1.5.2             htmltools_0.5.5          later_1.3.1              geneplotter_1.76.0      
[109] tidyr_1.3.0              DBI_1.1.3                MASS_7.3-60              ShortRead_1.54.0        
[113] Matrix_1.5-4.1           cli_3.6.1                parallel_4.2.1           metapod_1.4.0           
[117] igraph_1.4.3             pkgconfig_2.0.3          GenomicAlignments_1.34.0 numDeriv_2016.8-1.1     
[121] sp_1.6-0                 plotly_4.10.1            spatstat.sparse_3.0-1    annotate_1.76.0         
[125] XVector_0.38.0           stringr_1.5.0            digest_0.6.31            sctransform_0.3.5       
[129] RcppAnnoy_0.0.20         spatstat.data_3.0-1      Biostrings_2.66.0        rmarkdown_2.21          
[133] leiden_0.4.3             uwot_0.1.14              edgeR_3.38.4             restfulr_0.0.15         
[137] shiny_1.7.4              Rsamtools_2.14.0         gtools_3.9.4             rjson_0.2.21            
[141] lifecycle_1.0.3          nlme_3.1-162             jsonlite_1.8.4           viridisLite_0.4.2       
[145] limma_3.52.4             BSgenome_1.66.1          fansi_1.0.4              pillar_1.9.0            
[149] lattice_0.21-8           KEGGREST_1.38.0          fastmap_1.1.1            httr_1.4.6              
[153] survival_3.5-5           glue_1.6.2               png_0.1-8                bit_4.0.5               
[157] stringi_1.7.12           blob_1.2.4               DESeq2_1.38.1            memoise_2.0.1           
[161] latticeExtra_0.6-30      caTools_1.18.2           dplyr_1.1.2              irlba_2.3.5.1           
[165] future.apply_1.11.0

I notice that your version of DiffBind is a bit out of date -- current version is DiffBind_3.10, although this is unlikely to be the issue here.

I don't think that using loess offsets is the right way to go here as it tends to way over-normalize and eliminate the biological signal unless there is a specific technical trend in the data you need to correct.

It would be good to check the original MA plot using bNormalized=FALSE to verify that the skew towards more open chromatin in the control holds in the raw data. I assume you were using the default normalization based on library sizes? Are there big differences in sequencing depth? You may want to try normalizing using background bins (calling dba.normalize() with background=TRUE) but I doubt that this signal could be coming from the library normalization by itself.

I understand that you weren't expecting a loss of open chromatin in the FRDA group, but it may well be that case that this is what the data are showing.