Jim, the big guru has spoken some time ago and he said that tblastx is a very sensitive alignment tool, more than blastn or blat. You cannot run it on the whole genome, but you might give it a try on smaller sequences. However, blastToPsl cannot parse tblastx-output, as tblastx does not return real alignments (not sure here, but there seem to exist variable sized gaps "*" in the output), and without psl -> no chain, no chain -> no liftover. Sigh.
We can still simply take everything that overlaps an alignable region and map it to the WHOLE alignable region on the second sequence so as to give at least a hint where it should be roughly located. This is what the following scripts does. It needs the bed file from sequence1 and the result from tblastx with option "-m 9" (tabular format), aligning sequence1 on sequence2. The script will then project the bed file from sequence1 to the BEST hit from sequence2 and annotate the length of the intersection between the alignable region on sequence1 and the original annotation on sequence1.
This is handy to annotate non-coding exons for model species with bad gene models (e.g. tetraodon, fugu). Get the human sequence and exons as bed, then lift the human annotation to the tetraodon sequence. Any other idea how to do this?
Note: If you don't like tblastx because of its speed, you can replace it with yass which seems to be faster and just as (or, according to its paper) more sensitive than blastn. With -d 3 yass creates output identical to blast-tubular.
#!/usr/bin/python # written by Max, 8/06, in a damn night train, who can sleep in night trains? from sys import * from optparse import OptionParser # === COMMAND LINE INTERFACE, OPTIONS AND HELP === parser = OptionParser("%prog [options] bedfile blastfile: Use a tblastx-tabular-file to lift annotations between sequences (approximatively), very slow") (options, args) = parser.parse_args() # ==== FUNCTIONs ===== class hit: """ a hit from a blast tabular output file """ def __repr__(self): return self.line def parseBlast(file): hits =  for l in file: if l.startswith("#"): continue fs = l.split() if len(fs)!=12: stderr.write("error: blast output does not have 12 fields. Make sure that you ran blast with -m 9 or -D T (for bl2seq)\n") exit(1) h = hit() (h.qName, h.sName, h.percId, h.alnLen, h.mismatches, h.gapOpen, h.qStart, h.qEnd, h.sStart, h.sEnd, h.eVal, h.alnScore) = fs h.sStart = int(h.sStart) h.sEnd = int(h.sEnd) h.qStart = int(h.qStart) h.qEnd = int(h.qEnd) if h.qEnd < h.qStart: (h.qStart, h.qEnd) = (h.qEnd, h.qStart) if h.sEnd < h.sStart: (h.sStart, h.sEnd) = (h.sEnd, h.sStart) h.line = l.strip() hits.append(h) return hits class bed: """ a feature from a bed file """ def overlaps(self, hit): """ returns number of overlapping bp if two Features overlap or false if no overlap """ if (( hit.sStart <= self.start and hit.sEnd > self.start) or \ (hit.sStart < self.end and hit.sEnd >= self.end) or (hit.sStart > self.start and hit.sStart < self.end)): return (min (hit.sEnd, self.end) - max (hit.sStart, self.start)) else: return 0 def __repr__(self): return self.line def parseBed(file): beds =  for l in file: l = l.strip() b = bed() if l.startswith("#"): continue fs = l.split() if not len(fs) >= 4: stderr.write("error: bed features need names.\n") exit(1) (b.chrom, b.start, b.end, b.name) = fs[0:4] b.start = int(b.start) b.end = int(b.end) b.line = l beds.append(b) return beds # ----------- MAIN -------------- if args==: parser.print_help() exit(1) stderr.write("Reading blast output...\n") f = open(args, "r") hits = parseBlast(f) stderr.write("Reading bedfile ...\n") f = open(args, "r") beds = parseBed(f) stderr.write("Searching for overlaps bed-blast...\n") for b in beds: overlaps =  for h in hits: length = b.overlaps(h) if length > 0: h.overlapLen=length overlaps.append(h) # choose highest scoring overlap if len(overlaps)==0: continue overlaps.sort(key=lambda x: x.eVal, reverse=True) #for o in overlaps: #print o best = overlaps print best.qName, best.qStart, best.qEnd, b.name+"_ints"+str(best.overlapLen)