Hi Damien,
Thanks for providing a sample PDB file that illustrates the problem you are facing; it helped me to identify the issue.
As shown by the attached image for DT8 on chain F, the base is distorted beyond recognition. For example, C5--C6 distance is only 0.42 Å -- far too short for a covalent bond, whereas N1--C6 = 1.97 Å and C4--C5 = 2.16 Å are far too long. So 3DNA won't recognize it as a DNA base T at all. Same issues exist for the other three nucleotides.
It does not makes much sense to change 3DNA to accommodate such erroneous cases; rather, the mistakes should be fixed in any tool you used to generate this structure in the first place.
HTH,
Xiang-Jun