Because a full set of colors is required to account for color terms, it is not possible to reduce data for stars that have not been observed in all filters. If you have some stars that were observed in only a subset of bands, you should extract from the entire table of observations just the subset of bands, and run the whole set of stars in this band subset to obtain reduced values for the stars with incomplete observations. Then make a second set of data from which the incomplete observations are removed, and reduce these to get good values for the stars with complete observations. Be aware that the reduced values will differ systematically in the two subsets.
For example, some stars might be observed only in B and V in a run when most stars were observed in the full UBV set of bands. You would then do two separate reductions: one for all stars, using only the B and V data, and one with only the stars having full UBV data. Notice that the B and V values for the stars in common will differ slightly in the two solutions. The values from the full UBV solution should be more accurate; but you should not intermix them with values from the BV solution. If homogeneity is more important than accuracy, you could try adopting the B and V values for all stars from the BV solution, and the U-B colors from the full solution. The B's in the B-V colors then differ from the B's in the U-B colors; but this is basically what Johnson did in setting up the system. To avoid the problem, make sure you observe every star in every band.
Although the reduction program assumes you will observe a contiguous subset of the bands in any standard system, you could still force it to reduce a non-contiguous subset by replying OTHER when asked for the system name. You would then supply, in order of increasing wavelength, the bands you actually used. However, you would also have to make up a special set of standard-star files, in which the indices skip any missing bands.
Note that standard-star data that are missing some passbands may still be useful. In general, at least one magnitude is required; except for H-Beta standards, stars with only indices and no magnitudes are not useful.