aboutsummaryrefslogtreecommitdiff
path: root/wqflask
diff options
context:
space:
mode:
authorAlexander_Kabui2022-07-22 13:00:25 +0300
committerBonfaceKilz2022-07-25 22:20:46 +0300
commit43c67d5eaa5aa8ad7117d70b0b097696c639f618 (patch)
treefd0cd43dc5af4297fe1e0bada79566a777c0c03d /wqflask
parent9132ae308d1e4d3fbfc39e2fdf6c2e951b27c529 (diff)
downloadgenenetwork2-43c67d5eaa5aa8ad7117d70b0b097696c639f618.tar.gz
new file module for preprocessing input
Diffstat (limited to 'wqflask')
-rw-r--r--wqflask/wqflask/correlation/rust_correlation.py57
1 files changed, 57 insertions, 0 deletions
diff --git a/wqflask/wqflask/correlation/rust_correlation.py b/wqflask/wqflask/correlation/rust_correlation.py
new file mode 100644
index 00000000..ebb079fa
--- /dev/null
+++ b/wqflask/wqflask/correlation/rust_correlation.py
@@ -0,0 +1,57 @@
+"""module contains integration code for rust-gn3"""
+import json
+from wqflask.correlation.correlation_functions import get_trait_symbol_and_tissue_values
+from wqflask.correlation.correlation_gn3_api import create_target_this_trait
+from gn3.computations.rust_correlation import run_correlation
+from gn3.computations.rust_correlation import get_sample_corr_data
+from gn3.computations.rust_correlation import parse_tissue_corr_data
+
+
+def compute_correlation_rust(start_vars: dict, corr_type: str,
+ method: str = "pearson", n_top: int = 500):
+ """function to compute correlation"""
+
+ (this_dataset, this_trait, target_dataset,
+ sample_data) = create_target_this_trait(start_vars)
+
+ if corr_type == "sample":
+
+ all_samples = json.loads(start_vars["sample_vals"])
+ sample_data = get_sample_corr_data(sample_type=start_vars["corr_samples_group"],
+ all_samples=all_samples,
+ dataset_samples=this_dataset.group.all_samples_ordered())
+
+ target_dataset.get_trait_data(list(sample_data.keys()))
+
+ target_data = []
+ for (key, val) in target_dataset.trait_data.items():
+ lts = [key] + [str(x) for x in val]
+ r = ",".join(lts)
+ target_data.append(r)
+ # breakpoint()
+
+ results = run_correlation(target_data, ",".join(
+ [str(x) for x in list(sample_data.values())]), method, ",")
+
+
+ if corr_type == "tissue":
+
+ trait_symbol_dict = this_dataset.retrieve_genes("Symbol")
+ corr_result_tissue_vals_dict = get_trait_symbol_and_tissue_values(
+ symbol_list=list(trait_symbol_dict.values()))
+
+ data = parse_tissue_corr_data(symbol_name=this_trait.symbol,
+ symbol_dict=get_trait_symbol_and_tissue_values(
+ symbol_list=[this_trait.symbol]),
+ dataset_symbols=trait_symbol_dict,
+ dataset_vals=corr_result_tissue_vals_dict)
+
+ if data:
+ results = run_correlation(
+ data[1], data[0], method, ",")
+
+ return {"correlation_results": results[0:n_top],
+ "this_trait": this_trait.name,
+ "target_dataset": start_vars['corr_dataset'],
+ "return_results": n_top
+ }