Hello , I'm working on a idea of building a hidden markov model from synthetic de novo protein designs.

Particularly, I'm building a hidden markov model using hmmer3. The background is I made 200 synthetic structural designs of enzymes with RFDiffusion2 from a theozyme motif (5 aminoacids) previously described in a crystalized protein. From those protein designs of equal length (450aa), I ran ligandmnn to predict the aminoacidic sequence of my enzymatic designs, after corroborating appropriate folding with alphafold3, I got 25 self-consistent sequences in which the position of the 5 catalytic residues are in correct place (same as in the rfdiffusion2 models). So when I perform a multiple sequence alignment of these synthetic sequences with Clustal Omega, the 5 catalytic residues are not aligned in the same columns as I expect them to be. What workaround you recommend me to do to achieve/force the placement of these 5 essential aminoacids in the same positions in the multiple sequence alignment ?

By discussing with some colleges, one told me that neighbor aminoacids are important to consider in this approach. Any ideas?