diff tools/rgenetics/rgGLM.py @ 0:9071e359b9a3

Uploaded
author xuebing
date Fri, 09 Mar 2012 19:37:19 -0500
parents
children
line wrap: on
line diff
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/tools/rgenetics/rgGLM.py	Fri Mar 09 19:37:19 2012 -0500
@@ -0,0 +1,287 @@
+#!/usr/local/bin/python
+"""
+# added most of the available options for linear models
+# june 2009 rml
+# hack to run and process a plink quantitative trait
+#
+
+This is a wrapper for Shaun Purcell's Plink linear/logistic models for
+traits, covariates and genotypes.
+
+It requires some judgement to interpret the findings
+We need some better visualizations - manhattan plots are good.
+svg with rs numbers for top 1%?
+
+toptable tools - truncate a gg file down to some low percentile
+
+intersect with other tables - eg gene expression regressions on snps
+
+
+
+"""
+
+import sys,math,shutil,subprocess,os,string,tempfile,shutil,commands
+from rgutils import plinke
+
+def makeGFF(resf='',outfname='',logf=None,twd='.',name='track name',description='track description',topn=1000):
+    """
+    score must be scaled to 0-1000
+    
+    Want to make some wig tracks from each analysis
+    Best n -log10(p). Make top hit the window.
+    we use our tab output which has
+    rs	chrom	offset	ADD_stat	ADD_p	ADD_log10p
+    rs3094315	1	792429	1.151	0.2528	0.597223
+
+    """
+
+    def is_number(s):
+        try:
+            float(s)
+            return True
+        except ValueError:
+            return False
+    header = 'track name=%s description="%s" visibility=2 useScore=1 color=0,60,120\n' % (name,description)          
+    column_names = [ 'Seqname', 'Source', 'Feature', 'Start', 'End', 'Score', 'Strand', 'Frame', 'Group' ]
+    halfwidth=100
+    resfpath = os.path.join(twd,resf)
+    resf = open(resfpath,'r')
+    resfl = resf.readlines() # dumb but convenient for millions of rows
+    resfl = [x.split() for x in resfl]
+    headl = resfl[0]
+    resfl = resfl[1:]
+    headl = [x.strip().upper() for x in headl]
+    headIndex = dict(zip(headl,range(0,len(headl))))
+    chrpos = headIndex.get('CHROM',None)
+    rspos = headIndex.get('RS',None)
+    offspos = headIndex.get('OFFSET',None)
+    ppos = headIndex.get('ADD_LOG10P',None)
+    wewant = [chrpos,rspos,offspos,ppos]
+    if None in wewant: # missing something
+       logf.write('### Error missing a required header in makeGFF - headIndex=%s\n' % headIndex)
+       return
+    resfl = [x for x in resfl if x[ppos] > '']
+    resfl = [(float(x[ppos]),x) for x in resfl] # decorate
+    resfl.sort()
+    resfl.reverse() # using -log10 so larger is better
+    resfl = resfl[:topn] # truncate
+    pvals = [x[0] for x in resfl] # need to scale
+    resfl = [x[1] for x in resfl] # drop decoration
+    if len(pvals) == 0:
+        logf.write('### no pvalues found in resfl - %s' % (resfl[:3]))
+        sys.exit(1)
+    maxp = max(pvals) # need to scale
+    minp = min(pvals)
+    prange = abs(maxp-minp) + 0.5 # fudge
+    scalefact = 1000.0/prange
+    logf.write('###maxp=%f,minp=%f,prange=%f,scalefact=%f\n' % (maxp,minp,prange,scalefact))
+    for i,row in enumerate(resfl):
+        row[ppos] = '%d' % (int(scalefact*pvals[i]))
+        resfl[i] = row # replace
+    outf = file(outfname,'w')
+    outf.write(header)
+    outres = [] # need to resort into chrom offset order
+    for i,lrow in enumerate(resfl):
+        chrom,snp,offset,p, = [lrow[x] for x in wewant]
+        gff = ('chr%s' % chrom,'rgGLM','variation','%d' % (int(offset)-halfwidth),
+               '%d' % (int(offset)+halfwidth),p,'.','.','%s logp=%1.2f' % (snp,pvals[i]))
+        outres.append(gff)
+    outres = [(x[0],int(x[3]),x) for x in outres] # decorate
+    outres.sort() # into chrom offset
+    outres=[x[2] for x in outres] # undecorate
+    outres = ['\t'.join(x) for x in outres]    
+    outf.write('\n'.join(outres))
+    outf.write('\n')
+    outf.close()
+
+
+
+def xformQassoc(resf='',outfname='',logf=None,twd='.'):
+    """	plink.assoc.linear to gg file
+from the docs
+The output per each SNP might look something like:
+
+    CHR        SNP      BP  A1       TEST   NMISS       OR      STAT         P
+      5   rs000001   10001   A        ADD     664   0.7806    -1.942   0.05216
+      5   rs000001   10001   A     DOMDEV     664   0.9395   -0.3562    0.7217
+      5   rs000001   10001   A       COV1     664   0.9723   -0.7894    0.4299
+      5   rs000001   10001   A       COV2     664    1.159    0.5132    0.6078
+      5   rs000001   10001   A   GENO_2DF     664       NA     5.059    0.0797   
+    need to transform into gg columns for each distinct test
+    or bed for tracks?
+    
+    """
+    logf.write('xformQassoc got resf=%s, outfname=%s\n' % (resf,outfname))
+    resdict = {}
+    rsdict = {}
+    markerlist = []
+    # plink is "clever" - will run logistic if only 2 categories such as gender
+    resfs = resf.split('.')
+    if resfs[-1] == 'logistic':
+        resfs[-1] = 'linear'
+    else:
+        resfs[-1] = 'logistic'
+    altresf = '.'.join(resfs)
+
+    altresfpath = os.path.join(twd,altresf)
+    resfpath = os.path.join(twd,resf)
+    try:
+        resf = open(resfpath,'r')
+    except:
+        try:
+            resf = open(altresfpath,'r')
+        except:
+            print >> sys.stderr, '## error - no file plink output %s or %s found - cannot continue' % (resfpath, altresfpath)
+            sys.exit(1)
+    for lnum,row in enumerate(resf):
+        if lnum == 0:
+            headl = row.split()
+            headl = [x.strip().upper() for x in headl]
+            headIndex = dict(zip(headl,range(0,len(headl))))
+            chrpos = headIndex.get('CHR',None)
+            rspos = headIndex.get('SNP',None)
+            offspos = headIndex.get('BP',None)
+            nmisspos = headIndex.get('NMISS',None)
+            testpos = headIndex.get('TEST',None)
+            ppos = headIndex.get('P',None)
+            coeffpos = headIndex.get('OR',None)
+            if not coeffpos:
+                coeffpos = headIndex.get('BETA',None)
+            apos = headIndex.get('A1',None)
+            statpos = headIndex.get('STAT',None)
+            wewant = [chrpos,rspos,offspos,testpos,statpos,ppos,coeffpos,apos]
+            if None in wewant: # missing something
+               logf.write('missing a required header in xformQassoc - headIndex=%s\n' % headIndex)
+               return
+            llen = len(headl)        
+        else: # no Nones!
+            ll = row.split()
+            if len(ll) >= llen: # valid line
+                chrom,snp,offset,test,stat,p,coeff,allele = [ll[x] for x in wewant]
+                snp = snp.strip()
+                if p <> 'NA' :
+                  try:
+                    ffp = float(p)
+                    if ffp <> 0:
+                       lp =  -math.log10(ffp)
+                  except:
+                    lp = 0.0
+                  resdict.setdefault(test,{})
+                  resdict[test][snp] = (stat,p,'%f' % lp)
+                  if rsdict.get(snp,None) == None:
+                      rsdict[snp] = (chrom,offset)
+                      markerlist.append(snp)
+    # now have various tests indexed by rs
+    tk = resdict.keys()
+    tk.sort() # tests
+    ohead = ['rs','chrom','offset']
+    for t in tk: # add headers
+        ohead.append('%s_stat' % t)
+        ohead.append('%s_p' % t)
+        ohead.append('%s_log10p' % t)
+    oheads = '\t'.join(ohead)
+    res = [oheads,]
+    for snp in markerlist: # retain original order
+        chrom,offset = rsdict[snp]
+        outl = [snp,chrom,offset]
+        for t in tk:
+            outl += resdict[t][snp] # add stat,p for this test
+        outs = '\t'.join(outl)
+        res.append(outs)
+    f = file(outfname,'w')
+    res.append('')
+    f.write('\n'.join(res))
+    f.close()
+
+                
+if __name__ == "__main__":
+    """
+
+    <command interpreter="python">   
+        rgGLM.py '$i.extra_files_path/$i.metadata.base_name' '$phef.extra_files_path/$phef.metadata.base_name'
+        "$title1" '$predvar' '$covar' '$out_file1' '$logf' '$i.metadata.base_name'
+        '$inter' '$cond' '$gender' '$mind' '$geno' '$maf' '$logistic' '$wigout'
+    </command>
+    """
+    topn = 1000
+    killme = string.punctuation+string.whitespace
+    trantab = string.maketrans(killme,'_'*len(killme))
+    if len(sys.argv) < 17:
+       s = 'rgGLM.py needs 17 params - got %s \n' % (sys.argv)
+       sys.stderr.write(s) # print >>,s would probably also work?
+       sys.exit(0)
+    blurb = 'rgGLM.py called with %s' % sys.argv
+    print >> sys.stdout,blurb
+    bfname = sys.argv[1]
+    phename = sys.argv[2]
+    title = sys.argv[3]
+    title.translate(trantab)
+    predvar = sys.argv[4]
+    covar = sys.argv[5].strip()
+    outfname = sys.argv[6] 
+    logfname = sys.argv[7]
+    op = os.path.split(logfname)[0]
+    try: # for test - needs this done
+        os.makedirs(op)
+    except:
+        pass    
+    basename = sys.argv[8].translate(trantab)
+    inter = sys.argv[9] == '1'
+    cond = sys.argv[10].strip()
+    if cond == 'None':
+        cond = ''
+    gender = sys.argv[11] == '1'
+    mind = sys.argv[12]
+    geno = sys.argv[13]
+    maf = sys.argv[14]
+    logistic = sys.argv[15].strip()=='1'
+    gffout = sys.argv[16]
+    me = sys.argv[0]
+    phepath = '%s.pphe' % phename
+    twd = tempfile.mkdtemp(suffix='rgGLM') # make sure plink doesn't spew log file into the root!
+    tplog = os.path.join(twd,'%s.log' % basename) # should be path to plink log
+    vcl = [plinke,'--noweb','--bfile',bfname,'--pheno-name','"%s"' % predvar,'--pheno',
+           phepath,'--out',basename,'--mind %s' % mind, '--geno %s' % geno,
+           '--maf %s' % maf]
+    if logistic:
+        vcl.append('--logistic')
+        resf = '%s.assoc.logistic' % basename # plink output is here we hope
+    else:
+        vcl.append('--linear')
+        resf = '%s.assoc.linear' % basename # plink output is here we hope
+    resf = os.path.join(twd,resf)
+    if gender:
+        vcl.append('--sex')
+    if inter:
+        vcl.append('--interaction')
+    if covar > 'None':
+        vcl += ['--covar',phepath,'--covar-name',covar] # comma sep list of covariates
+    tcfile = None
+    if len(cond) > 0: # plink wants these in a file..
+        dummy,tcfile = tempfile.mkstemp(suffix='condlist') #
+        f = open(tcfile,'w')
+        cl = cond.split()
+        f.write('\n'.join(cl))
+        f.write('\n')
+        f.close()
+        vcl.append('--condition-list %s' % tcfile)
+    p=subprocess.Popen(' '.join(vcl),shell=True,cwd=twd)
+    retval = p.wait()
+    if tcfile:
+        os.unlink(tcfile)
+    plinklog = file(tplog,'r').read()
+    logf = file(logfname,'w')
+    logf.write(blurb)
+    logf.write('\n')
+    logf.write('vcl=%s\n' % vcl)
+    xformQassoc(resf=resf,outfname=outfname,logf=logf,twd=twd) # leaves the desired summary file
+    makeGFF(resf=outfname,outfname=gffout,logf=logf,twd=twd,name='rgGLM_TopTable',description=title,topn=topn)
+    logf.write('\n')
+    logf.write(plinklog)
+    logf.close()
+    #shutil.rmtree(twd) # clean up
+
+
+
+
+