To be safe, I would just take the counts from the tximport list and
cbind() them with your other counts matrix. Basically, you are limited in your analysis options because you only have counts for one dataset. There is not the option to use the methods described and recommended in the tximport paper (length offsets, or counts-from-abundance). You can use
tximport to import the counts, but just use
txi$counts. And then you are just using observed counts in both datasets with no length offset. This is often not a problem (as we note in the paper), but the other tximport methods protect against bias from changes in the effective length of the feature across the samples.