def sortbysign(expdat,field=False,value='',exclude=False,exact=True,maxfval=0.2):
"""
sort bacteria in the experiment based on the number of positive/negative samples
(ignoring nans)
input:
expdat : Experiment
field,value,exclude,exact : name of field and value of field in order to sort based only on these samples
or field=False for all samples (default)
maxfval - the maximal f-value
output:
newexp : Experiment
sorted by difference between positive/negative
"""
params=locals()
if field:
texp=hs.filtersamples(expdat,field,value,exact=exact,exclude=exclude)
else:
texp=hs.copyexp(expdat)
texp.data=np.sign(texp.data)
numpos=np.nansum(texp.data>0,axis=1)
numneg=np.nansum(texp.data<0,axis=1)
pval=np.ones(len(numpos))
for cpos in range(len(pval)):
if numpos[cpos]==0 and numneg[cpos]==0:
continue
pval1=stats.binom.cdf(numpos[cpos],numpos[cpos]+numneg[cpos],0.5)
pval2=stats.binom.cdf(numneg[cpos],numpos[cpos]+numneg[cpos],0.5)
pval[cpos]=np.min([pval1,pval2])
signs=np.nanmean(texp.data,axis=1)
fval=hs.fdr(pval)
keep=np.where(np.array(fval)<=maxfval)[0]
newexp=hs.reorderbacteria(expdat,keep)
signs=signs[keep]
si=np.argsort(signs)
newexp=hs.reorderbacteria(newexp,si)
newexp.filters.append("sort by sign field %s max-f-val %f" % (field,maxfval))
hs.addcommand(newexp,"sortbysign",params=params,replaceparams={'expdat':expdat})
return newexp
def annotationenrichment(expdat,seqs,compareseqs=None,fdrval=0.1):
"""
get a list of annotations enriched in seqs compared to random draw from expdat
input:
expdat : Experiment
seqs : list of sequences ('ACGT')
the sequences in which to test the enrichment
compareseqs : list of sequences ('ACGT')
the sequences to compare to
None (default) to compare to all the experiment
output:
newplist - a sorted list of dict for annotaions which are below fdr ('description','pval','observed','expected')
"""
# if annotations not initialized - get them (to save time)
if expdat.seqannotations is None:
expdat=hs.getexpannotations(expdat,usesupercooldb=False)
if expdat.annotationseqs is None:
expdat=hs.getexpannotations(expdat,usesupercooldb=False)
# count the number of annotations for each term in the group
# into a dict {term:total number observed in group}
groupannotationcount=collections.defaultdict(int)
totgroup=0
for cseq in seqs:
for cinfo in expdat.seqannotations[cseq]:
groupannotationcount[cinfo]+=1
totgroup+=1
# count the number of annotations for each term in the comparison group
# into a dict {term:total number observed in comparison group}
if compareseqs is None:
compareseqs=expdat.seqs
compgroupannotationcount=collections.defaultdict(int)
totcompgroup=0
for cseq in compareseqs:
for cinfo in expdat.seqannotations[cseq]:
compgroupannotationcount[cinfo]+=1
totcompgroup+=1
hs.Debug(6,'%d annotations in group, %d in all' % (totgroup,totcompgroup))
# calculate the probability per term
# note: we use a bad calculation (can choose the same term twice for a single bacteria). need to improve (permutations?)
pvals={}
allp=[]
pv=[]
for cinfo in expdat.annotationseqs.keys():
pcompgroup=float(compgroupannotationcount[cinfo])/totcompgroup
pval1=stats.binom.cdf(groupannotationcount[cinfo],totgroup,pcompgroup)
pval2=stats.binom.cdf(totgroup-groupannotationcount[cinfo],totgroup,1-pcompgroup)
p=np.min([pval1,pval2])
# p=pval1
pvals[cinfo]=p
allp.append(p)
cpv={}
cpv['pval']=p
cpv['observed']=groupannotationcount[cinfo]
cpv['expected']=pcompgroup*totgroup
cpv['description']=cinfo
pv.append(cpv)
fval=hs.fdr(allp)
keep=np.where(np.array(fval)<=fdrval)
plist=[]
rat=[]
for cidx in keep[0]:
plist.append(pv[cidx])
rat.append(np.abs(float(pv[cidx]['observed']-pv[cidx]['expected']))/np.mean([pv[cidx]['observed'],pv[cidx]['expected']]))
si=np.argsort(rat)
si=si[::-1]
newplist=[]
for idx,crat in enumerate(rat):
newplist.append(plist[si[idx]])
for cp in newplist:
if cp['observed']>cp['expected']:
hs.Debug(6,cp)
return(newplist)