def condition_inference(M, who_hits, exposure_hits, who_Mitocheck, num_permutations, threshold, random_result,
taking_siRNAs=False):
'''
- M: distance matrix of size (hits, mitocheck)
- taking_siRNAs: indicates if you want to consider different values of the same siRNAs independently (False) or as replicates of the same condition
'''
r={}
yqualdict=expSi('../data/mapping_2014/qc_export.txt')
yqualdict.update(expSi('../data/mapping_2014/qc_validation_exp.txt', primary_screen=False))
dictSiEntrez=siEntrez('../data/mapping_2014/mitocheck_siRNAs_target_genes_Ens75.txt')
siRNAs=[yqualdict[e] for e in who_Mitocheck]
genes=[dictSiEntrez[e] for e in siRNAs]
count=Counter(exposure_hits)
#This way I look at exposures that are hits at least 50% of the times/plates
for el in filter(lambda x: count[x]/float(PASSED_QC_COND[x])>0.5, count):
print el
where_=np.where(exposure_hits==el)[0]
batch_names = who_hits[where_]
curr_dist=np.hstack((M[j] for j in where_))
curr_who=[(batch_names[0], '') for k in range(M.shape[1])]
curr_conditions=list(who_Mitocheck) if not taking_siRNAs else list(siRNAs)
for name in batch_names[1:]:
curr_who.extend([(name, '') for k in range(M.shape[1])])
if not taking_siRNAs:
curr_conditions.extend(who_Mitocheck)
else:
curr_conditions.extend(siRNAs)
print curr_dist.shape
r[el], random_result=computeRPpvalues(curr_dist, np.array(curr_who), conditions=np.array(curr_conditions), technical_replicates_key=np.median,
num_permutations=num_permutations, reverse=False,
batch_names=batch_names, random_result=random_result,
signed=False)
r[el]=sorted(r[el], key=itemgetter(2))
pval=np.array(np.array(r[el])[:,-1], dtype=float)
if not taking_siRNAs:
currG=[dictSiEntrez[yqualdict[e]] for e in np.array(r[el])[np.where(pval<=threshold)][:,0]]
else:
currG=[dictSiEntrez[e] for e in np.array(r[el])[np.where(pval<=threshold)][:,0]]
print sorted(Counter(currG).keys())
return r
def condition_cluster_inference(M, clusters, who_hits, exposure_hits, who_Mitocheck, num_permutations, threshold, random_result, filename,
taking_siRNAs=True, gsea=False):
'''
- M: distance matrix of size (hits, mitocheck)
- taking_siRNAs: indicates if you want to consider different values of the same siRNAs independently (False) or as replicates of the same condition
- num_permutations: no calculation of p-values if None, else number of permutations
- filename if we want to write GSEA ranking files
- gsea : if you want to write gsea ranking files
'''
r={}
yqualdict=expSi('../data/mapping_2014/qc_export.txt')
yqualdict.update(expSi('../data/mapping_2014/qc_validation_exp.txt', primary_screen=False))
dictSiEntrez=siEntrez('../data/mapping_2014/mitocheck_siRNAs_target_genes_Ens75.txt')
trad=EnsemblEntrezTrad('../data/mapping_2014/mitocheck_siRNAs_target_genes_Ens75.txt')
siRNAs=[yqualdict[e] for e in who_Mitocheck]
genes=[dictSiEntrez[e] for e in siRNAs]
past_cluster_num=0
for cluster_num in clusters:
print cluster_num
r[cluster_num]={'conditions':clusters[cluster_num]}
where_=np.hstack((np.where(exposure_hits==el)[0] for el in clusters[cluster_num]))
batch_names = who_hits[where_]
curr_dist=np.hstack((M[j] for j in where_))
curr_who=[(batch_names[0], '') for k in range(M.shape[1])]
curr_conditions=list(who_Mitocheck) if not taking_siRNAs else list(genes)
for name in batch_names[1:]:
curr_who.extend([(name, '') for k in range(M.shape[1])])
if not taking_siRNAs:
curr_conditions.extend(who_Mitocheck)
else:
curr_conditions.extend(genes)
print curr_dist.shape
if num_permutations is not None and past_cluster_num!=len(batch_names):
#If there are only five experiments that are different between the two cluster length, then no need to redo the random rank product computation
random_result=None
curr_res=computeRPpvalues(curr_dist, np.array(curr_who),
conditions=np.array(curr_conditions),
technical_replicates_key=np.median,
xb_screen=False,
num_permutations=num_permutations, reverse=False,
batch_names=batch_names, random_result=random_result,
signed=False)
#donc curr_res est [(gene, rank value)]
if num_permutations is None and gsea:
writeGSEARankingFile(curr_res, filename.format(cluster_num))
else:
if len(curr_res)==2:
#this means that we have the p-values
curr_res, random_result=curr_res
curr_res=sorted(curr_res, key=itemgetter(-1))
pval=np.array(np.array(curr_res)[:,-1], dtype=float)
if not taking_siRNAs:
currG=[dictSiEntrez[yqualdict[e]] for e in np.array(curr_res)[np.where(pval<=threshold)][:,0]]
else:
currG=[e for e in np.array(curr_res)[np.where(pval<=threshold)][:,0]]
print sorted(currG)
else:
#this means that we're working with the rank product values
curr_res=sorted(curr_res, key=itemgetter(-1))
if not taking_siRNAs:
currG=[dictSiEntrez[yqualdict[e]] for e in np.array(curr_res)[:threshold][:,0]]
else:
currG=[e for e in np.array(curr_res)[:threshold][:,0]]
print sorted(currG)
past_cluster_num=len(batch_names)
r[cluster_num]['genes']=[trad[el] for el in currG]
r[cluster_num]['result']=[(el[0],el[-1]) for el in curr_res]
multipleGeneListsToFile([r[k]['genes'] for k in r], [k for k in r], name=filename)
return r
