while using unsupervised learning, what to do when the trained weight vector size donot match the new input size?

I am using unsupervised learning ART on protein sequences. I have a trained network but now the weight matrix and the new input protein sequence (accessed from pdb) are of different size. which seems very natural but in this case how can i proceed to prove the model's (ART's) efficiency? I mean what changes in the regular algorithm of update weights , should I do to get desired results? plz help.