diff options
Diffstat (limited to 'wqflask')
-rw-r--r-- | wqflask/wqflask/show_trait/show_trait.py | 23 |
1 files changed, 13 insertions, 10 deletions
diff --git a/wqflask/wqflask/show_trait/show_trait.py b/wqflask/wqflask/show_trait/show_trait.py index f1bd6f27..30b03e66 100644 --- a/wqflask/wqflask/show_trait/show_trait.py +++ b/wqflask/wqflask/show_trait/show_trait.py @@ -43,8 +43,6 @@ logger = getLogger(__name__ ) class ShowTrait(object): def __init__(self, kw): - logger.debug("in ShowTrait, kw are:", kw) - if 'trait_id' in kw and kw['dataset'] != "Temp": self.temp_trait = False self.trait_id = kw['trait_id'] @@ -152,6 +150,7 @@ class ShowTrait(object): #self.nearest_marker1 = "" #self.nearest_marker2 = "" + self.make_sample_lists() self.qnorm_vals = quantile_normalize_vals(self.sample_groups) @@ -397,6 +396,7 @@ class ShowTrait(object): return_results_menu_selected = return_results_menu_selected,) def make_sample_lists(self): + all_samples_ordered = self.dataset.group.all_samples_ordered() parent_f1_samples = [] @@ -405,13 +405,18 @@ class ShowTrait(object): primary_sample_names = list(all_samples_ordered) + if not self.temp_trait: other_sample_names = [] - for sample in self.this_trait.data.keys(): - if (self.this_trait.data[sample].name2 in primary_sample_names) and (self.this_trait.data[sample].name not in primary_sample_names): - primary_sample_names.append(self.this_trait.data[sample].name) - primary_sample_names.remove(self.this_trait.data[sample].name2) - elif sample not in all_samples_ordered: + + for sample in self.this_trait.data: + if self.this_trait.data[sample].name2 != self.this_trait.data[sample].name: + if (self.this_trait.data[sample].name2 in primary_sample_names) and (self.this_trait.data[sample].name not in primary_sample_names): + primary_sample_names.append(self.this_trait.data[sample].name) + primary_sample_names.remove(self.this_trait.data[sample].name2) + + all_samples_set = set(all_samples_ordered) + if sample not in all_samples_set: all_samples_ordered.append(sample) other_sample_names.append(sample) @@ -424,6 +429,7 @@ class ShowTrait(object): primary_header = "%s Only" % (self.dataset.group.name) else: primary_header = "Samples" + primary_samples = SampleList(dataset = self.dataset, sample_names=primary_sample_names, this_trait=self.this_trait, @@ -523,9 +529,7 @@ def get_z_scores(sample_groups): def get_nearest_marker(this_trait, this_db): this_chr = this_trait.locus_chr - logger.debug("this_chr:", this_chr) this_mb = this_trait.locus_mb - logger.debug("this_mb:", this_mb) #One option is to take flanking markers, another is to take the two (or one) closest query = """SELECT Geno.Name FROM Geno, GenoXRef, GenoFreeze @@ -536,7 +540,6 @@ def get_nearest_marker(this_trait, this_db): ORDER BY ABS( Geno.Mb - {}) LIMIT 1""".format(this_chr, this_db.group.name+"Geno", this_mb) logger.sql(query) result = g.db.execute(query).fetchall() - logger.debug("result:", result) if result == []: return "" |