Single-sample Gene Set Enrichment Analysis (ssGSEA)¶

ssGSEA calculates pathway enrichment scores for each individual sample, enabling per-sample assessment of biological pathway activity. In synthetic data evaluation, preserving the distribution of these enrichment scores tests whether generation methods capture the complex correlation structures underlying individual sample heterogeneity.

KS-Complement Score¶

Preservation is measured using the KS-Complement score, which quantifies how well the distribution of Normalized Enrichment Scores (NES) is preserved across all 50 Hallmark pathways:

\[\text{KSComplement}_p = 1 - \text{KS}\!\bigl(\text{CDF}(\text{NES}_p^{\text{original}}),\;\text{CDF}(\text{NES}_p^{\text{synthetic}})\bigr)\]

The overall performance for a given method is the mean KS-Complement score across all pathways.

Code Example¶

The ssGSEA evaluation reuses the UnivariateSimilarity class from synomicsbench.metrics.fidelity to compare NES distributions pathway-by-pathway:

from synomicsbench.metrics.fidelity.UnivariateSimilarity import UnivariateSimilarity
from synomicsbench.processing.metadata import MetaData
import pandas as pd

# Load ssGSEA NES matrices (rows = samples, columns = pathways)
ori_ssgsea = pd.read_csv("ssGSEA_Origin.csv")
df_pivot_ori = ori_ssgsea.pivot(index="Name", columns="Term", values="NES")

syn_ssgsea = pd.read_csv("ssGSEA_gaussiancopula_0.csv")
df_pivot_syn = syn_ssgsea.pivot(index="Name", columns="Term", values="NES")

# Auto-detect metadata (all pathways are numerical)
metadata = MetaData.get_metadata(
    data=df_pivot_ori,
    threshold_unique_values=10,
    ordinal_features=None
)

# Compute KS-Complement scores across all pathways
uni = UnivariateSimilarity(output_dir="results/ssgsea")
scores = uni.get_univariate_score(
    original_data=df_pivot_ori,
    synthetic_data=df_pivot_syn,
    metadata=metadata
)
detail_df = uni.get_detail_df()

print(f"Mean KS-Complement Score: {scores:.4f}")

The df_pivot_ori and df_pivot_syn objects are DataFrames where rows represent samples and columns correspond to Hallmark pathways:

Patient	HALLMARK_ADIPOGENESIS	HALLMARK_ALLOGRAFT_REJECTION	HALLMARK_ANDROGEN_RESPONSE
Patient1	0.3660	0.1872	0.3924
Patient2	0.3632	0.2673	0.4165
Patient3	0.4251	0.2897	0.4295
Patient4	0.3856	0.2036	0.4002
Patient5	0.3489	0.3721	0.3289