In the Dx.R output file, for a sample I'm looking at, the column "somatic_ontarget" is 0 in 119 entries and 1 in 2 entries. What is this column? It sounds like it should be a count of number of somatic variants that were found to lie within the Dx.R target areas?
Yes. Have a look at section 10.4 and Table 5 for a description of the output.
If there are no somatic mutations in almost all files, then there is most likely an issue with the VCFs.
Check the predictSomatic output (Sampleinfo_variants.csv). If the somatic variants are missing, then they are removed by the QC filters (see the log file for details). If they are assigned a low prior probability being somatic, your somatic annotation in the VCF is wrong. If they are all classified as germline despite being clearly somatic and labeled as such, then there is something very wrong with the setup. If they show up properly in predictSomatic output, then you filter them out in Dx.R using a wrong CallableLoci file for example.
I would recommend Mutect 1.1.7 for test runs.