def run_ssgsea(filepath: str, output_dir: str):
"""Perform ssGSEA."""
# Read data
gene_exp = pd.read_csv(filepath, sep='\t')
kegg_gene_set, _, _, _ = check_gmt_files()
# Filter data set such that only those genes which are in the gene sets are in the expression data
filtered_expression_data = filter_gene_exp_data(gene_exp, merge_gene_set)
logger.info(f'Running {filepath}')
single_sample_gsea = gseapy.ssgsea(
data=filtered_expression_data,
gene_sets=kegg_gene_set,
outdir=output_dir, # do not write output to disk
sample_norm_method='rank', # choose 'custom' for your own rank list
permutation_num=
0, # skip permutation procedure, because you don't need it
no_plot=True, # skip plotting to speed up
format='png',
)
logger.info('Done with ssGSEA')
single_sample_gsea.res2d.to_csv(os.path.join(output_dir, 'results.tsv'),
sep='\t')
def do_ssgsea(
filtered_expression_data: pd.DataFrame,
gene_set: str,
output_dir: str = None,
processes: int = 96,
max_size: int = 3000,
min_size: int = 15,
) -> SingleSampleGSEA:
"""Run single sample GSEA (ssGSEA) on filtered gene expression data set.
:param filtered_expression_data: filtered gene expression values for samples
:param gene_set: .gmt file containing gene sets
:param output_dir: output directory
:param processes: Number of processes
:param max_size: Maximum allowed number of genes from gene set also the data set
:param min_size: Minimum allowed number of genes from gene set also the data set
:return: ssGSEA results in respective directory
"""
single_sample_gsea = ssgsea(
data=filtered_expression_data,
gene_sets=gene_set,
outdir=output_dir, # do not write output to disk
max_size=max_size,
min_size=min_size,
sample_norm_method='rank', # choose 'custom' for your own rank list
permutation_num=0, # skip permutation procedure, because you don't need it
no_plot=True, # skip plotting to speed up
processes=processes,
format='png',
)
return single_sample_gsea
def run_ssgsea(
filtered_expression_data: pd.DataFrame,
gene_set: str,
output_dir: str = SSGSEA,
processes: int = 1,
max_size: int = 3000,
min_size: int = 15,
) -> SingleSampleGSEA:
"""Run single sample GSEA (ssGSEA) on filtered gene expression data set.
:param filtered_expression_data: filtered gene expression values for samples
:param gene_set: .gmt file containing gene sets
:param output_dir: output directory
:return: ssGSEA results in respective directory
"""
single_sample_gsea = gseapy.ssgsea(
data=filtered_expression_data,
gene_sets=gene_set,
outdir=output_dir, # do not write output to disk
max_size=max_size,
min_size=min_size,
sample_norm_method='rank', # choose 'custom' for your own rank list
permutation_num=
0, # skip permutation procedure, because you don't need it
no_plot=True, # skip plotting to speed up
processes=processes,
format='png',
)
logger.info('Done with ssGSEA')
return single_sample_gsea
def pathway_enrichment(self,
factor,
views=None,
genesets=None,
nprocesses=4,
permutation_num=0):
if genesets is None:
genesets = [
"c6.all.v7.1.symbols.gmt",
"c5.all.v7.1.symbols.gmt",
"h.all.v7.1.symbols.gmt",
"c2.all.v7.1.symbols.gmt",
]
if views is None:
views = ["methylation", "transcriptomics", "proteomics"]
df = pd.concat(
[
gseapy.ssgsea(
self.weights[v][factor],
processes=nprocesses,
permutation_num=permutation_num,
gene_sets=Enrichment.read_gmt(f"{DPATH}/pathways/{g}"),
no_plot=True,
).res2d.assign(geneset=g).assign(view=v).reset_index()
for v in views for g in genesets
],
ignore_index=True,
)
df = df.rename(columns={"sample1": "nes"}).sort_values("nes")
return df
def run_ssGSEA(self, ordered_df, db,
method='rank', no_plot=True,
processes=4):
ssgsea_analysis = gseapy.ssgsea(ordered_df,
gene_sets=db,
outdir='ss_'+self.gsea_save,
sample_norm_method=method,
no_plot=no_plot,
processes=processes)
return ssgsea_analysis
def main():
# Import data
df, tar, grp, lengths, _ = data_reader.read_main(raw=True)
# Log transform keeping all columns as they will be used in the gsea.
df = df_utils.transform_sequence_to_microarray(df.T, all=True)
# Handling for microarray set 0 as this does not require log transformation
"""
df1, _, _ = data_reader.read_number(1)
df2, _, _ = data_reader.read_number(2)
df_len = len(df)
df = df_utils.merge_frames([df,df1,df2], drop=False)
df = df.head(df_len)
"""
sample = df.T
ss = gseapy.ssgsea(data=sample
, gene_sets='Out/gmt_hepmark.gmt'
, no_plot=True
, outdir='Out/gsea_hepmark'
, min_size=10)
# "When you run the gene set enrichment analysis, the GSEA software automatically normalizes
# the enrichment scores for variation in gene set size, as described in GSEA Statistics.
# Nevertheless, the normalization is not very accurate for extremely small or extremely
# large gene sets. For example, for gene sets with fewer than 10 genes, just 2 or 3 genes
# can generate significant results. Therefore, by default, GSEA ignores gene sets that
# contain fewer than 25 genes or more than 500 genes; defaults that are appropriate for
# datasets with 10,000 to 20,000 features. To change these default values, use the Max Size
# and Min Size parameters on the Run GSEA Page; however, keep in mind the possibility of
# inflated scorings for very small gene sets and inaccurate normalization for large ones."
# Setup df file
rows = []
for s in ss.resultsOnSamples:
row = [s]
for val in ss.resultsOnSamples[s]:
row.append(val)
rows.append(row)
columns = ['index']
columns.extend([x for x in ss.resultsOnSamples[s].axes[0]])
# NB : Do not use the res2d as this is the normalized score
# Create es file
df_out = pd.DataFrame(rows, columns = columns)
df_out.set_index('index', inplace=True)
path = r'%s' % getcwd().replace('\\','/') + "/Out/enrichment_scores/"
df_out.to_csv(path+"es_test.csv")
("prot_culture_reps", "PC1"),
("prot_culture_reps", "PC2"),
("prot_broad_culture", "PC1"),
("prot_culture_reps_emt", "PC1"),
("prot_culture_reps_emt", "PC2"),
("prot_broad_culture_emt", "PC1"),
("prot_broad_culture_emt", "PC2"),
("prot_broad_culture_emt", "PC4"),
]
enr_pcs = pd.concat(
[
gseapy.ssgsea(
dsets_dred[dtype]["loadings"].loc[dtype_pc],
processes=4,
gene_sets=Enrichment.read_gmt(f"{DPATH}/pathways/{g}"),
no_plot=True,
).res2d.assign(geneset=g).assign(dtype=dtype).assign(
dtype_pc=dtype_pc).reset_index() for dtype, dtype_pc in enr_pcs
for g in genesets
],
ignore_index=True,
)
enr_pcs = enr_pcs.rename(columns={"sample1": "nes"}).sort_values("nes")
enr_pcs.to_csv(f"{RPATH}/DimRed_pcs_enr.csv.gz",
compression="gzip",
index=False)
# Plot
enr_pcs_plt = [
("prot", "PC1", "prot_broad", "PC1", 0.5),
), time.ctime()
# gene sets for ssGSEA
gene_sets = {}
pw_list = []
for pw in pathwayDic[testing_pathway.lower()]:
for uniprot in pathwayDic[testing_pathway.lower()][pw]:
if uniprot in uniprot2gene:
gene = uniprot2gene[uniprot]
if gene in geneList:
if not pw in gene_sets:
gene_sets[pw] = []
gene_sets[pw].append(gene)
pw_list = gene_sets.keys()
fo = open('%s/%s.gmt' % (fo_directory, testing_pathway.lower()), 'w')
for pw in pw_list:
print >> fo, pw + '\t' + '\t'.join(gene_sets[pw])
fo.close()
# ssGSEA
ss = gp.ssgsea(data='%s/expression.txt' % (fo_directory),
outdir='%s/%s_ssgsea_result' %
(fo_directory, testing_pathway.lower()),
gene_sets='%s/%s.gmt' %
(fo_directory, testing_pathway.lower()),
sample_norm_method='rank',
permutation_num=0,
no_plot=True,
scale=True,
min_size=2)
def ssgsea(self, name: str, df: pd.DataFrame,
gene_sets: dict) -> pd.DataFrame:
"""
Perform single sample GSEA.
Args:
name: str, name of analysis, used to create directory and store results.
df: pd.DataFrame, the data
gene_sets: dict, a dict of gene sets for which to perform the analysis.
Structure is (gene set name) -> Path(...)
Returns:
DataFrame with all the samples NES and a column indicating the gene set name.
"""
dfs = []
for name_gs, val_gs in gene_sets.items():
outdir_gs = Enricher.ENRICHMENT_DIR_PATH / name
if not outdir_gs.is_dir():
outdir_gs.mkdir()
if isinstance(val_gs, Path):
geneset_dict = read_gmt(val_gs)
else:
geneset_dict = val_gs.copy()
# Perform the enrichment in batches of size step_size
genesets = list(geneset_dict.keys())
step_size = 40
dfs_geneset = []
for i in range(0, len(genesets) + step_size, step_size):
# Define genesets_batch
if i == 0:
continue
if i < len(genesets):
genesets_batch = genesets[i - step_size:i]
else:
genesets_batch = genesets[i - step_size:len(genesets)]
genesets_batch_dict = {
geneset: geneset_dict[geneset]
for geneset in genesets_batch
}
print(i - step_size, i, len(genesets_batch_dict))
ss = gp.ssgsea(data=df,
gene_sets=genesets_batch_dict,
outdir=str(outdir_gs / name_gs),
sample_norm_method=self.sample_norm_method,
min_size=self.min_size,
max_size=self.max_size,
permutation_num=self.permutation_num,
scale=self.scale,
processes=self.processes,
no_plot=self.no_plot,
verbose=self.verbose)
df_res_batch = pd.DataFrame.from_dict(
ss.resultsOnSamples, orient='index').transpose()
dfs_geneset.append(df_res_batch)
# Aggregate batches
df_res = pd.concat(dfs_geneset, axis=0, sort=False)
df_res['gene_set'] = name_gs
df_res.to_csv(outdir_gs / (name_gs + ".csv"),
header=True,
index=True)
dfs.append(df_res)
# Aggregate the different categories of gene sets (e.g. KEGG, GO, etc.)
df_agg = pd.concat(dfs, axis=0, sort=False)
df_agg.to_csv(outdir_gs / "enrichment_results.csv",
header=True,
index=True)
return df_agg