"""module contains code for correlations""" from typing import List from typing import Tuple from typing import Optional from typing import Callable import scipy.stats def compute_sum(rhs: int, lhs: int) -> int: """Initial tests to compute sum of two numbers""" return rhs + lhs def map_shared_keys_to_values(target_sample_keys: List, target_sample_vals: dict)-> List: """Function to construct target dataset data items given commoned shared\ keys and trait samplelist values for example given keys >>>>>>>>>>\ ["BXD1", "BXD2", "BXD5", "BXD6", "BXD8", "BXD9"] and value object as\ "HCMA:_AT": [4.1, 5.6, 3.2, 1.1, 4.4, 2.2],TXD_AT": [6.2, 5.7, 3.6, 1.5, 4.2, 2.3]}\ return results should be a list of dicts mapping the shared keys to the trait values""" target_dataset_data = [] for trait_id, sample_values in target_sample_vals.items(): target_trait_dict = dict(zip(target_sample_keys, sample_values)) target_trait = { "trait_id": trait_id, "trait_sample_data": target_trait_dict } target_dataset_data.append(target_trait) return target_dataset_data def normalize_values(a_values: List, b_values: List) -> Tuple[List[float], List[float], int]: """Trim two lists of values to contain only the values they both share Given two lists of sample values, trim each list so that it contains only the samples that contain a value in both lists. Also returns the number of such samples. >>> normalize_values([2.3, None, None, 3.2, 4.1, 5], [3.4, 7.2, 1.3, None, 6.2, 4.1]) ([2.3, 4.1, 5], [3.4, 6.2, 4.1], 3) """ a_new = [] b_new = [] for a_val, b_val in zip(a_values, b_values): if (a_val and b_val is not None): a_new.append(a_val) b_new.append(b_val) return a_new, b_new, len(a_new) def compute_corr_coeff_p_value(primary_values: List, target_values: List, corr_method: str) -> Tuple[float, float]: """Given array like inputs calculate the primary and target_value methods -> pearson,spearman and biweight mid correlation return value is rho and p_value """ corr_mapping = { "bicor": do_bicor, "pearson": scipy.stats.pearsonr, "spearman": scipy.stats.spearmanr } use_corr_method = corr_mapping.get(corr_method, "spearman") corr_coeffient, p_val = use_corr_method(primary_values, target_values) return (corr_coeffient, p_val) def compute_sample_r_correlation( corr_method: str, trait_vals, target_samples_vals) -> Optional[Tuple[float, float, int]]: """Given a primary trait values and target trait values calculate the correlation coeff and p value """ (sanitized_traits_vals, sanitized_target_vals, num_overlap) = normalize_values(trait_vals, target_samples_vals) if num_overlap > 5: (corr_coeffient, p_value) =\ compute_corr_coeff_p_value(primary_values=sanitized_traits_vals, target_values=sanitized_target_vals, corr_method=corr_method) # xtodo check if corr_coefficient is None # should use numpy.isNan scipy.isNan is deprecated if corr_coeffient is not None: return (corr_coeffient, p_value, num_overlap) return None def do_bicor(x_val, y_val) -> Tuple[float, float]: """Not implemented method for doing biweight mid correlation use astropy stats package :not packaged in guix """ return (x_val, y_val) def filter_shared_sample_keys(this_samplelist, target_samplelist) -> Tuple[List, List]: """Given primary and target samplelist\ for two base and target trait select\ filter the values using the shared keys""" this_vals = [] target_vals = [] for key, value in target_samplelist.items(): if key in this_samplelist: target_vals.append(value) this_vals.append(this_samplelist[key]) return (this_vals, target_vals) def compute_all_sample_correlation(this_trait, target_dataset, corr_method="pearson") -> List: """Given a trait data samplelist and\ target__datasets compute all sample correlation """ this_trait_samples = this_trait["trait_sample_data"] corr_results = [] for target_trait in target_dataset: trait_id = target_trait.get("trait_id") target_trait_data = target_trait["trait_sample_data"] this_vals, target_vals = filter_shared_sample_keys( this_trait_samples, target_trait_data) sample_correlation = compute_sample_r_correlation( corr_method=corr_method, trait_vals=this_vals, target_samples_vals=target_vals) if sample_correlation is not None: (corr_coeffient, p_value, num_overlap) = sample_correlation else: continue corr_result = { "corr_coeffient": corr_coeffient, "p_value": p_value, "num_overlap": num_overlap } corr_results.append({trait_id: corr_result}) return corr_results def tissue_lit_corr_for_probe_type(corr_type: str, top_corr_results): """Function that does either lit_corr_for_trait_list or tissue_corr _for_trait list depending on whether both dataset and target_dataset are both set to probet """ corr_results = {"lit": 1} if corr_type not in ("lit", "literature"): corr_results["top_corr_results"] = top_corr_results # run lit_correlation for the given top_corr_results if corr_type == "tissue": # run lit correlation the given top corr results pass if corr_type == "sample": pass # run sample r correlation for the given top results return corr_results def tissue_correlation_for_trait_list( primary_tissue_vals: List, target_tissues_values: List, corr_method: str, compute_corr_p_value: Callable = compute_corr_coeff_p_value) -> dict: """Given a primary tissue values for a trait and the target tissues values compute the correlation_cooeff and p value the input required are arrays output -> List containing Dicts with corr_coefficient value,P_value and also the tissue numbers is len(primary) == len(target) """ # ax :todo assertion that length one one target tissue ==primary_tissue (tissue_corr_coeffient, p_value) = compute_corr_p_value(primary_values=primary_tissue_vals, target_values=target_tissues_values, corr_method=corr_method) lit_corr_result = { "tissue_corr": tissue_corr_coeffient, "p_value": p_value, "tissue_number": len(primary_tissue_vals) } return lit_corr_result def fetch_lit_correlation_data( conn, input_mouse_gene_id: Optional[str], gene_id: str, mouse_gene_id: Optional[str] = None) -> Tuple[str, float]: """Given input trait mouse gene id and mouse gene id fetch the lit\ corr_data""" if mouse_gene_id is not None and ";" not in mouse_gene_id: query = """ SELECT VALUE FROM LCorrRamin3 WHERE GeneId1='%s' and GeneId2='%s' """ query_values = (str(mouse_gene_id), str(input_mouse_gene_id)) cursor = conn.cursor() cursor.execute(query_formatter(query, *query_values)) results = cursor.fetchone() lit_corr_results = None if results is not None: lit_corr_results = results else: cursor = conn.cursor() cursor.execute(query_formatter(query, *tuple(reversed(query_values)))) lit_corr_results = cursor.fetchone() lit_results = (gene_id, lit_corr_results.val)\ if lit_corr_results else (gene_id, 0) return lit_results return (gene_id, 0) def lit_correlation_for_trait_list( conn, target_trait_lists: List, species: Optional[str] = None, trait_gene_id: Optional[str] = None) -> List: """given species,base trait gene id fetch the lit corr results from the db\ output is float for lit corr results """ fetched_lit_corr_results = [] this_trait_mouse_gene_id = map_to_mouse_gene_id(conn=conn, species=species, gene_id=trait_gene_id) for (trait_name, target_trait_gene_id) in target_trait_lists: corr_results = {} if target_trait_gene_id: target_mouse_gene_id = map_to_mouse_gene_id( conn=conn, species=species, gene_id=target_trait_gene_id) fetched_corr_data = fetch_lit_correlation_data( conn=conn, input_mouse_gene_id=this_trait_mouse_gene_id, gene_id=target_trait_gene_id, mouse_gene_id=target_mouse_gene_id) dict_results = dict(zip(("gene_id", "lit_corr"), fetched_corr_data)) corr_results[trait_name] = dict_results fetched_lit_corr_results.append(corr_results) return fetched_lit_corr_results def query_formatter(query_string: str, *query_values): """Formatter query string given the unformatted query string\ and the respectibe values.Assumes number of placeholders is equal to the number of query values """ # xtodo escape sql queries results = query_string % (query_values) return results def map_to_mouse_gene_id(conn, species: Optional[str], gene_id: Optional[str]) -> Optional[str]: """Given a species which is not mouse map the gene_id\ to respective mouse gene id""" # AK:xtodo move the code for checking nullity out of thing functions bug # while method for string if None in (species, gene_id): return None if species == "mouse": return gene_id cursor = conn.cursor() query = """SELECT mouse FROM GeneIDXRef WHERE '%s' = '%s'""" query_values = (species, gene_id) cursor.execute(query_formatter(query, *query_values)) results = cursor.fetchone() mouse_gene_id = results.mouse if results is not None else None return mouse_gene_id def compute_all_lit_correlation(conn, trait_lists: List, species: str, gene_id): """Function that acts as an abstraction for lit_correlation_for_trait_list""" lit_results = lit_correlation_for_trait_list( conn=conn, target_trait_lists=trait_lists, species=species, trait_gene_id=gene_id) return {"lit_results": lit_results} def compute_all_tissue_correlation(primary_tissue_dict: dict, target_tissues_data: dict, corr_method: str): """Function acts as an abstraction for tissue_correlation_for_trait_list\ required input are target tissue object and primary tissue trait\ target tissues data contains the trait_symbol_dict and symbol_tissue_vals """ tissues_results = {} primary_tissue_vals = primary_tissue_dict["tissue_values"] traits_symbol_dict = target_tissues_data["trait_symbol_dict"] symbol_tissue_vals_dict = target_tissues_data["symbol_tissue_vals_dict"] target_tissues_list = process_trait_symbol_dict( traits_symbol_dict, symbol_tissue_vals_dict) for target_tissue_obj in target_tissues_list: trait_id = target_tissue_obj.get("trait_id") target_tissue_vals = target_tissue_obj.get("tissue_values") tissue_result = tissue_correlation_for_trait_list( primary_tissue_vals=primary_tissue_vals, target_tissues_values=target_tissue_vals, corr_method=corr_method) tissues_results[trait_id] = tissue_result return tissues_results def process_trait_symbol_dict(trait_symbol_dict, symbol_tissue_vals_dict) -> List: """Method for processing trait symbol\ dict given the symbol tissue values """ traits_tissue_vals = [] for (trait, symbol) in trait_symbol_dict.items(): if symbol is not None: target_symbol = symbol.lower() if target_symbol in symbol_tissue_vals_dict: trait_tissue_val = symbol_tissue_vals_dict[target_symbol] target_tissue_dict = {"trait_id": trait, "symbol": target_symbol, "tissue_values": trait_tissue_val} traits_tissue_vals.append(target_tissue_dict) return traits_tissue_vals