def findmislabels(expdat,field,distmetric='bc'): """" find mislabelled samples according to field input: expdat : Experiment field : string name of the field to examine (i.e. subjectid) distmetric : string the distance meteric to use (see calcdist) """ expdat=hs.sortsamples(expdat,field) fvals=hs.getfieldvals(expdat,field) ufvals=list(set(fvals)) onames=[] for idx,csamp in enumerate(expdat.samples): onames.append(csamp+';'+fvals[idx]) omat=np.zeros([len(fvals),len(ufvals)]) for groupidx,groupval in enumerate(ufvals): cexp=hs.filtersamples(expdat,field,groupval,exact=True) for aidx,aval in enumerate(expdat.samples): cdist=[] for gidx,gval in enumerate(cexp.samples): # don't measure distance to ourselves if gval==aval: continue cdist.append(hs.calcdist(cexp.data[:,gidx],expdat.data[:,aidx],distmetric=distmetric)) omat[aidx,groupidx]=np.mean(cdist) figure() iax=imshow(omat,interpolation='nearest',aspect='auto') ax=iax.get_axes() ax.set_xticks(range(len(ufvals))) ax.set_xticklabels(ufvals,rotation=90) ax.set_yticks(range(len(onames))) ax.set_yticklabels(onames)
def sortbycentermass(expdat,field=False,numeric=True,uselog=True):
"""
sort bacteria in the experiment according to a 1d gradient by calculating the center of mass
input:
expdat
field : string
the name of the field to sort by or False to skip sorting
numeric : bool
True if the sort field is numeric (ignored if no sort field)
uselog : bool
True to log transform the data before mass center calculation
output:
newexp - the experiment with sorted bacteria
"""
params=locals()
if field:
newexp=hs.sortsamples(expdat,field,numeric=numeric)
else:
newexp=hs.copyexp(expdat)
dat=newexp.data
if uselog:
dat[dat<1]=1
dat=np.log2(dat)
cm=[]
multpos=np.arange(len(newexp.samples))
for cseqind in range(len(newexp.seqs)):
cm.append(np.dot(dat[cseqind,:],multpos)/np.sum(dat[cseqind,:]))
sv,si=hs.isort(cm)
newexp=hs.reorderbacteria(expdat,si)
newexp.filters.append("sort by center of mass field=%s, uselog=%s" % (field,uselog))
hs.addcommand(newexp,"sortbycentermass",params=params,replaceparams={'expdat':expdat})
return newexp
def sortsamples(self):
items=self.bMainList.selectedItems()
if len(items)!=1:
print("Need 1 item")
return
for citem in items:
cname=str(citem.text())
cexp=self.explist[cname]
sortsampleswin = SortSamplesWindow(cexp)
res=sortsampleswin.exec_()
if res==QtGui.QDialog.Accepted:
field=str(sortsampleswin.cField.currentText())
newname=str(sortsampleswin.tNewName.text())
cnumeric=sortsampleswin.cNumeric.checkState()
if cnumeric==0:
numeric=False
else:
numeric=True
overwrite=sortsampleswin.cOverwrite.checkState()
newexp=hs.sortsamples(cexp,field=field,numeric=numeric)
if overwrite==0:
newexp.studyname=newname
self.addexp(newexp)
else:
self.replaceexp(newexp)
def filterwave(expdat,field=False,numeric=True,minfold=2,minlen=3,step=1,direction='up',posloc='start'):
"""
filter bacteria, keeping only ones that show a consecutive region of samples with higher/lower mean than other samples
Done by scanning all windowlen/startpos options for each bacteria
input:
expdat : Experiment
field : string
The field to sort by or False to skip sorting
numeric : bool
For the sorting according to field (does not matter if field is False)
minfold : float
The minimal fold change for the window compared to the rest in order to keep
step : int
The skip between tested windows (to make it faster use a larger skip)
minlen : int
The minimal window len for over/under expression testing
direction : string
'both' - test both over and under expression in the window
'up' - only overexpressed
'down' - only underexpressed
posloc : string
The position to measure the beginning ('maxstart') or middle ('maxmid') of maximal wave
or 'gstart' to use beginning of first window with >=minfold change
output:
newexp : Experiment
The filtered experiment, sorted according to window start samples position
"""
params=locals()
# sort if needed
if field:
newexp=hs.sortsamples(expdat,field,numeric=numeric)
else:
newexp=hs.copyexp(expdat)
dat=newexp.data
dat[dat<1]=1
dat=np.log2(dat)
numsamples=len(newexp.samples)
numbact=len(newexp.seqs)
maxdiff=np.zeros([numbact])
maxpos=np.zeros([numbact])-1
maxlen=np.zeros([numbact])
for startpos in range(numsamples-minlen):
for cwin in np.arange(minlen,numsamples-startpos,step):
meanin=np.mean(dat[:,startpos:startpos+cwin],axis=1)
nowin=[]
if startpos>0:
nonwin=np.arange(startpos-1)
if startpos<numsamples:
nowin=np.hstack([nowin,np.arange(startpos,numsamples-1)])
nowin=nowin.astype(int)
meanout=np.mean(dat[:,nowin],axis=1)
cdiff=meanin-meanout
if direction=='both':
cdiff=np.abs(cdiff)
elif direction=='down':
cdiff=-cdiff
if posloc=='gstart':
usepos=np.logical_and(cdiff>=minfold,maxpos==-1)
maxpos[usepos]=startpos
elif posloc=='start':
maxpos[cdiff>maxdiff]=startpos
elif posloc=='mid':
maxpos[cdiff>maxdiff]=startpos+int(cwin/2)
else:
hs.Debug('posloc nut supported %s' % posloc)
return False
maxlen[cdiff>maxdiff]=cwin
maxdiff=np.maximum(maxdiff,cdiff)
keep=np.where(maxdiff>=minfold)[0]
keeppos=maxpos[keep]
si=np.argsort(keeppos)
keep=keep[si]
for ci in keep:
hs.Debug(6,'bacteria %s startpos %d len %d diff %f' % (newexp.tax[ci],maxpos[ci],maxlen[ci],maxdiff[ci]))
newexp=hs.reorderbacteria(newexp,keep)
newexp.filters.append('Filter wave field=%s minlen=%d' % (field,minlen))
hs.addcommand(newexp,"filterwave",params=params,replaceparams={'expdat':expdat})
hs.Debug(6,'%d samples before filtering, %d after' % (len(expdat.samples),len(newexp.samples)))
return newexp
