next up previous
Next: The neural network Up: Methods Previous: Alignments

Filtered sequence database

PSIBLAST is an iterative searching method. During each iteration, it is possible for the searching profile to become polluted with sequences that although show significant similarity to the query, ought not be included. This can be caused by low complexity sequence matching the query, or by matching sequences of biased composition. We applied SEG [48], to filter the search database, and so `masked out' regions of low complexity sequence. Coiled coil regions and transmembrane helices (TM's) were also masked out from the database. Masking these regions was performed using HELIXFILT. HELIXFILT looks for heptad repeats for coils and also uses the membrane potentials from MEMSTAT [49] to mask coils and transmembrane spans. HELIXFILT was kindly made available by Dr. D. Jones, and is also used as part of the PSIPRED [39] method.



James Cuff
2001-06-29