From SMian@lbl.gov Fri May 19 09:33:15 2000 Return-Path: Sender: saira@lbl.gov Date: Fri, 19 May 2000 09:33:11 -0700 From: Saira Mian X-Accept-Language: en To: karplus@cse.ucsc.edu Subject: General comments Content-Type: text/plain; charset=us-ascii Dear Kevin, A general strategy I've found works well with large numbers of fairly related sequences (as in, for example T90), is to start with an HMM that has a very high value for the cutmatch parameter (>0.95). This picks out highly conserved "motif blocks". Once defined, their match states are fixed (type definition A) and the model retrained with a smaller cutmatch to tidy up the intervening regions. One fairly easy approach for improving models/alignments is to use a pseudo Gibbs sampling method. Given a model, align all but the worst scoring sequence to the model and then use the a2m file and modelfromalign to reestimate a new HMM. Align all the sequences to the new HMM, this usually fixes some of the problems with the poor sequence. Repeat with other problematic seqences. Although the likelihood of the model may not change significantly, the alignment does often improve. -saira -- I. Saira Mian Life Sciences Division (Mail Stop 74-197) E-mail: SMian@lbl.gov Lawrence Berkeley National Laboratory Tel: (510) 486-6216 1 Cyclotron Road Fax: (510) 486-6949 Berkeley, California 94720