Hi,

I am running deseq2 on a data set with multiple factors and getting this warning:

the design formula contains a numeric variable with integer values,
  specifying a model with increasing fold change for higher values.
  did you mean for this to be a factor? if so, first convert
  this variable to a factor using the factor() function

What difference does it make for deseq2 if I am using quantitative variable (numerical values) instead of quantitative variables (factors)?

Will there be a difference in the end results? (I guess so, otherwise, there is no reason to put the warning, but I can't understand what)

thanks

Assa

> sessionInfo()
R version 3.2.3 (2015-12-10)
Platform: x86_64-apple-darwin13.4.0 (64-bit)
Running under: OS X 10.11.2 (El Capitan)

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] parallel  stats4    stats     graphics  grDevices utils     datasets  methods   base     

other attached packages:
 [1] readr_0.2.2                WriteXLS_4.0.0             BiocParallel_1.4.3         data.table_1.9.6          
 [5] hwriter_1.3.2              GOstats_2.36.0             graph_1.48.0               Category_2.36.0           
 [9] GO.db_3.2.2                AnnotationDbi_1.32.3       Matrix_1.2-3               ggplot2_2.0.0             
[13] gplots_2.17.0              biomaRt_2.26.1             ReportingTools_2.10.0      RSQLite_1.0.0             
[17] DBI_0.3.1                  knitr_1.12.3               RColorBrewer_1.1-2         genefilter_1.52.0         
[21] DESeq2_1.10.1              RcppArmadillo_0.6.400.2.2  Rcpp_0.12.3                SummarizedExperiment_1.0.2
[25] Biobase_2.30.0             GenomicRanges_1.22.3       GenomeInfoDb_1.6.3         IRanges_2.4.6             
[29] S4Vectors_0.8.7            BiocGenerics_0.16.1        stringr_1.0.0             

loaded via a namespace (and not attached):
 [1] edgeR_3.12.0             splines_3.2.3            R.utils_2.2.0            gtools_3.5.0             Formula_1.2-1           
 [6] highr_0.5.1              latticeExtra_0.6-26      RBGL_1.46.0              BSgenome_1.38.0          Rsamtools_1.22.0        
[11] lattice_0.20-33          biovizBase_1.18.0        limma_3.26.6             chron_2.3-47             XVector_0.10.0          
[16] colorspace_1.2-6         ggbio_1.18.3             R.oo_1.19.0              plyr_1.8.3               OrganismDbi_1.12.1      
[21] GSEABase_1.32.0          XML_3.98-1.3             zlibbioc_1.16.0          xtable_1.8-0             scales_0.3.0            
[26] gdata_2.17.0             annotate_1.48.0          PFAM.db_3.2.2            GenomicFeatures_1.22.11  nnet_7.3-11             
[31] survival_2.38-3          magrittr_1.5             evaluate_0.8             R.methodsS3_1.7.0        GGally_1.0.1            
[36] foreign_0.8-66           BiocInstaller_1.20.1     tools_3.2.3              formatR_1.2.1            munsell_0.4.2           
[41] locfit_1.5-9.1           cluster_2.0.3            lambda.r_1.1.7           Biostrings_2.38.3        caTools_1.17.1          
[46] futile.logger_1.4.1      grid_3.2.3               RCurl_1.95-4.7           dichromat_2.0-0          VariantAnnotation_1.16.4
[51] AnnotationForge_1.12.2   bitops_1.0-6             gtable_0.1.2             reshape_0.8.5            reshape2_1.4.1          
[56] GenomicAlignments_1.6.3  gridExtra_2.0.0          rtracklayer_1.30.1       Hmisc_3.17-1             futile.options_1.0.0    
[61] KernSmooth_2.23-15       stringi_1.0-1            geneplotter_1.48.0       rpart_4.1-10             acepack_1.3-3.3

"What difference does it make for deseq2 if I am using quantitative variable (numerical values) instead of quantitative variables (factors)? Will there be a difference in the end results?"

Yes there is a difference. 

This is actually just a "message". There are three levels in R: message, warning and error.

The point of the message is that, users may have a column, such as condition: 1,2,3,4. Converting this variable to a factor means that each of these groups is modeled with a separate term. This is a flexible and very useful way to do modeling, and this is the typical way that DESeq2 is used. If the condition was labeled A,B,C,D then this happens automatically.

It is also possible to have numeric covariates with DESeq2. But this means something very specific which the majority of users (I believe) do not intend. This means: there is a constant fold change for every unit of change in the condition. So if the estimated fold change is 2, this implies that condition 2 = 2x condition 1, condition 3 = 2x condition 2, etc. Or in other words, the relationship is linear on the log counts scale. 

It is a message instead of a warning, because some users may want to express such a relationship if the covariate is, e.g. titration of a chemical with a known log-linear relationship on gene expression, or something like this.

But even with numeric covariates like "age", it is preferred to use the more flexible modeling where you cut() the covariate into bins and fit a term for each bin. Please see the FAQ in the DESeq2 vignette about continuous variables.