Traditional transcriptomics relies heavily on comparison. To understand the molecular drivers of a disease, researchers typically compare a cohort of patient samples against a cohort of healthy controls. By calculating which genes are differentially expressed between these groups, standard Gene Set Enrichment Analysis (GSEA) identifies biological pathways that are systematically altered.
However, this cohort-based paradigm fails in clinical reality. A physician diagnosing a patient does not have a large, perfectly matched control cohort processed on the exact same sequencing run; they have exactly one sample from one individual. Because raw gene expression values are highly susceptible to platform noise and batch effects, absolute expression thresholds cannot be reliably applied across different laboratories. To achieve true personalized medicine, we need a way to quantify biological pathway activity within a single, isolated sample.
Single-sample pathway enrichment solves this problem by shifting the analytical frame of reference from the population to the individual transcriptome.
The Intuition of Internal Ranking
How do you calculate an enrichment score without a baseline group for comparison? The core mechanism relies on transforming absolute expression values into internal, relative ranks.
Instead of comparing the expression of Gene $X$ across fifty different patients, single-sample methods—such as single-sample Gene Set Enrichment Analysis (ssGSEA) and Gene Set Variation Analysis (GSVA)—rank every single gene within that individual sample based on its absolute expression level. If a sample expresses 20,000 genes, they are sorted from 1 (highest expression) to 20,000 (lowest expression).
To determine if a specific biological pathway (such as the TCA cycle or interferon signaling) is active, the algorithm evaluates whether the genes belonging to that pathway's predefined gene set are non-randomly distributed toward the top of the sample's internal ranking. By calculating an enrichment score based on these internal ranks, high-dimensional, noisy gene expression data is compressed into a stable, low-dimensional vector of pathway activity scores.
As demonstrated by Benjamin Haibe-Kains et al. in their development of the MM2S: personalized diagnosis of medulloblastoma patients and model systems classifier, this transformation eliminates the requirement for a matched healthy reference or a large processing batch. Because internal ranks preserve the relative biological signal while discarding absolute scale, the resulting pathway scores are highly robust against platform noise. This allows researchers to directly compare single-sample pathway profiles across entirely different datasets and tissue types, a strategy leveraged by Zhongxi Huang in Correlation Analysis Connects Cancer Subtypes to uncover shared biological traits across distinct cancers.
Measuring Functional Footprints Over Upstream Drivers
A parallel facet of single-sample pathway inference focuses on measuring downstream "functional footprints" rather than the upstream components themselves.
In clinical oncology, it is common to sequence a tumor to find mutations or over-expression in specific receptors, such as EGFR or PI3K. However, Andreas Mock et al. show in EGFR and PI3K Pathway Activities Might Guide Drug Repurposing in HPV-Negative Head and Neck Cancers that genetic mutations explain only a small fraction of actual downstream pathway signaling. A receptor might be mutated but inactive, or unmutated but hyper-activated by alternative cellular crosstalk.
Instead of measuring the receptor, footprint-based methods infer pathway activity by evaluating the collective expression of hundreds of downstream target genes known to be transcribed when the pathway is active. This concept is also central to the work of Elena Fountzilas et al. in A microRNA activity map of human mesenchymal tumors: connections to oncogenic pathways; an integrative transcriptomic study, where microRNA regulatory activity is mapped not by measuring the microRNAs themselves, but by modeling the coordinated behavior of their target genes using probabilistic regression. By focusing on the downstream transcriptomic footprint, these methods capture the actual functional output of a pathway within an individual sample.
Pathways as Features for Machine Learning
Once a single-sample transcriptome is converted into a set of pathway activity scores, these scores can be treated as continuous, robust features for downstream statistical modeling and machine learning.
Because pathway scores aggregate information from dozens or hundreds of genes, they are far less sparse and variable than individual gene measurements. This mathematical stability makes them ideal for training predictive models. For example:
- Continuous Phenotyping: Dvir Aran et al. utilized ssGSEA in Widespread parainflammation in human cancer to calculate a continuous score for a low-grade inflammatory state across individual primary tumors, allowing them to establish a clear link between this genomic signature and p53 mutations.
- Machine Learning Classifiers: Zhen Ren et al. applied GSVA in Machine learning-driven classification and prognostic prediction of kidney renal clear cell carcinoma using APOBEC family expression signatures to transform gene-level data into non-parametric enrichment scores, which were then fed into machine learning frameworks to stratify patients into distinct risk groups.
- Network-Based Integration: Silpa Suthram et al. integrated individual expression data with protein interaction networks in Network-Based Elucidation of Human Disease Similarities Reveals Common Functional Modules Enriched for Pluripotent Drug Targets to calculate collective activity metrics for functional modules, mapping hidden similarities across 54 human diseases.
Clinical Stratification and Drug Repurposing
The ultimate utility of single-sample pathway quantification lies in its power to guide personalized clinical decisions. By translating high-dimensional genomic data into interpretable biological processes, researchers can construct targeted prognostic indices and identify therapeutic vulnerabilities unique to an individual patient's tumor.
Prognostic Biomarkers
By scoring specific biological programs sample-by-sample, researchers can map the heterogeneity of complex cellular states across patient cohorts. This approach has been widely adopted to build specialized risk scores:
- Metabolic Heterogeneity: Qiang Zhang et al. used ssGSEA in An extensive bioinformatics study on the role of mitochondrial solute carrier family 25 in PC and its mechanism behind affecting immune infiltration and tumor energy metabolism to quantify individual metabolic activity, mapping how mitochondrial solute carriers dictate tumor energy metabolism and immune evasion.
- Cellular Stemness: Deli Mao et al. leveraged single-sample scoring in Identification of Stemness Characteristics Associated With the Immune Microenvironment and Prognosis in Gastric Cancer to quantify the degree to which individual gastric cancer samples retained undifferentiated, self-renewing properties.
- Programmed Cell Death: Shuai Ma et al. developed a "Necroptosis Index" using ssGSEA in Systematic analysis of the necroptosis index in pan-cancer and classification in discriminating the prognosis and immunotherapy responses of 1716 glioma patients to measure programmed necrotic cell death, successfully identifying patient subgroups highly responsive to immunotherapy.
- Microenvironmental Transitions: Similar single-sample strategies have been used to quantify epithelial-mesenchymal transitions (Jiasheng Wu et al. in Identification of EMT-Related Genes and Prognostic Signature With Significant Implications on Biological Properties and Oncology Treatment of Lower Grade Gliomas), cell adhesion dynamics (Xu Zheng et al. in Pan-cancer and machine learning reveal the role of cell adhesion molecules in prognosis, immune remodeling, and anti-tumour therapy in lower-grade gliomas), and structural gene family activation (Zhengtian Li et al. in Immunological role and prognostic value of the SKA family in pan-cancer analysis).
Therapeutic Alignment
Beyond prognosis, single-sample pathway profiles enable precise therapeutic alignment. In Prioritization of candidate cancer drugs based on a drug functional similarity network constructed by integrating pathway activities and drug activities, Jieyi Di et al. demonstrated that integrating individual ssGSEA pathway activity profiles with drug activity databases yields significantly higher accuracy in prioritizing candidate drugs for repurposing than traditional gene-expression matching.
By evaluating the collective behavior of gene sets within a single sample, these methods bypass the noise of individual transcripts, providing a reliable, interpretable, and clinically actionable map of an individual's biology.
Key Foundational Work
Network-Based Elucidation of Human Disease Similarities Reveals Common Functional Modules Enriched for Pluripotent Drug Targets (2010) — contributed 1 concept elements to this direction.
Identifying dysregulated pathways in cancers from pathway interaction networks (2012) — contributed 1 concept elements to this direction.
A microRNA activity map of human mesenchymal tumors: connections to oncogenic pathways; an integrative transcriptomic study (2012) — contributed 1 concept elements to this direction.