from collections import Counter
import numpy as np
import pandas as pd
[docs]def mc_gene_panel_greedy_search(cluster_cef_path,
pwdmg_path,
cef_value_ascending,
n_cef=500,
final_genes=None,
min_dmg_support=2,
selection_target=100,
add_gene_step=10):
if final_genes is None:
final_genes = {}
else:
# genes are preselected
if not isinstance(final_genes, dict):
final_genes = {g: 'pre-selected' for g in final_genes}
# only select top n_cef genes for each cluster, when the cluster is involved in PWDMG
cef = pd.read_csv(cluster_cef_path, index_col=0)
selected_cef = {col: cef[col].dropna().sort_values(ascending=cef_value_ascending)[:n_cef].index
for col in cef.columns}
# for each cluster pairs, save all significant DMGs for selection
pwdmg = pd.read_csv(pwdmg_path, index_col=0)
selected_genes = {}
for (a, b), sub_df in pwdmg.groupby(['hypo_in', 'hyper_in']):
possible_gene = sub_df.loc[sub_df.index.isin(selected_cef[a])].index
selected_genes[(a, b)] = possible_gene
# greedy search genes satisfy most PWDMG pairs, until any of these conditions achieved
# 1. all pairs have > min_dmg_support genes included
# 2. the selection_target gene number reached
# 3. there is no more significant PWDMG & CEF remained
final_genes, unsatisfied_pairs = dict_greedy_search(
final_dict=final_genes,
source_dict=selected_genes,
min_value_support=min_dmg_support,
selection_target=selection_target,
add_gene_step=add_gene_step)
# print final info after selection
print(f'{len(final_genes)} genes in final list.')
if len(unsatisfied_pairs) == 0:
print(f'All cluster pairs satisfied by having at least {min_dmg_support} DMGs')
else:
print(f'{len(unsatisfied_pairs)} cluster pairs unsatisfied.')
print(unsatisfied_pairs)
for c, i in pd.Series(np.concatenate(unsatisfied_pairs)).value_counts().items():
print(f'Cluster {c} in {i} pairs')
return final_genes
[docs]def dict_greedy_search(source_dict,
final_dict: dict,
min_value_support,
selection_target,
add_gene_step):
dmg_rank = 0
while True:
unsatisfied_keys = []
cur_vote = Counter()
for key, genes in source_dict.items():
if genes.isin(final_dict).sum() >= min_value_support:
continue
else:
cur_vote += Counter(genes)
unsatisfied_keys.append(key)
if (len(cur_vote) > 0) and (len(final_dict) < selection_target):
_this_step = min(add_gene_step, selection_target - len(final_dict))
ordered_genes = pd.Series(cur_vote).sort_values(ascending=False).index
genes_to_add = 0
reasons = {}
for g in ordered_genes:
if g in final_dict:
continue
genes_to_add += 1
if genes_to_add > _this_step:
break
reasons[g] = f'Greedy-{dmg_rank}'
dmg_rank += 1
if genes_to_add < _this_step:
# no enough genes to add
print('no enough genes to add')
break
final_dict.update(reasons)
else:
break
return final_dict, unsatisfied_keys
