Any unrepresented symbols at a given position may actually be found in new sequences, so a small weighting (0.025) towards these symbols is included in the table, unless the /STRINGENT option is used. If we have a large enough family creating an alignment the effect of including an unrepresented symbol is much smaller.
Instead of the usual symbol comparison tables used by GAP, BESTFIT etc, we use the new profile table. We can compare single sequences to the profile (PROFILEGAP), or scan a database (PROFILESEARCH). The comparison allows a new sequence to be aligned optimally to the family of sequences. The comparison of a sequence to a profile can emphasise similarity to conserved regions while tolerating diversity in other regions.
For a full description of the profiling method you should read the essay on "Profile analysis" by John Devereux in chapter 7 of the GCG program manual.