I have a control set of samples, and 3 viruses, each exposed to two treatments. 5 replicates of each. In two viruses, in both treatments, the first two replicates are left shifted in the PC1 values compared to the other replicates, example:

• virus1_treatment1_Rep1 -26.21084691
• virus1_treatment1_Rep2 -23.81517717
• virus1_treatment1_Rep3 -13.52695811
• virus1_treatment1_Rep4 -10.74887635
• virus1_treatment1_Rep5 -10.11061144

but in the controls, and the third virus, all 5 replicates look the same. I have results using a design of ~ virus_treatments + replicate, but I was wondering if that was right when the replicate effect is so different between the different viruses.

Is there a better way to deal with this?

Not really, the whole point is that the replicates account for within group variability. You can always access quality of a sample through independent measures, and then exclude on objective criterion.