> For the complete documentation index, see [llms.txt](https://docs.owkin.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.owkin.com/explore-and-analyse-data/understand-your-data/analysis-methods.md).
# Analysis methods
A catalogue of the analyses K-Pro can run on your data, the statistical methods behind them, and which datasets support each modality.
K-Pro runs published, peer-reviewed statistical methods on patient-level data across seven modalities: bulk RNA sequencing, whole exome sequencing (WES), single-cell RNA sequencing, spatial transcriptomics, H\&E histomics, clinical variables, and multimodal combinations of these.
This page catalogues **what each analysis does and when to use it**. For the plots these analyses produce — with parameters, screenshots, and example prompts — see **Example: Visualisations by modalities**.
## Bulk RNA sequencing
Bulk RNA-seq measures average gene expression across all cells in a sample, giving a robust and cost-effective basis for transcriptomic profiling, biomarker discovery, and molecular subtyping.
### Differential expression
* **Differential expression (DESeq2)**: identifies genes expressed differently between two clinical groups using the DESeq2 negative binomial model. Supports covariate adjustment for confounders such as age, gender, or tumour stage.
* **Differential expression (t-test)**: a modality-agnostic alternative using Welch's t-test, suitable for normalized expression data or when DESeq2 assumptions do not hold.
### Survival and prognosis
* **Survival analysis (Cox)**: Cox proportional hazards modelling of the effect of gene expression on overall survival (OS) or progression-free survival (PFS). Supports univariate models, multivariate models combining genes with clinical covariates, and interaction analyses across expression-defined subgroups.
* **Kaplan-Meier analysis**: survival curves stratified by expression level (for example high vs. low on a median cutoff) or another grouping variable.
* **Covariate-adjusted survival**: Kaplan-Meier-style curves that control for clinical covariates, isolating the independent effect of expression on survival.
### Pathways and signatures
* **ssGSEA enrichment**: single-sample gene set enrichment, scoring each sample for pathway or signature activity so groups can be compared without a predefined contrast.
* **GSEA (preranked)**: gene set enrichment on a ranked gene list, such as the output of a differential expression analysis, returning a normalized enrichment score per pathway.
* **Gene signature scoring**: custom signature scores across samples, for comparing biological programs such as immune signatures, metabolic pathways, or stemness scores.
### Clustering and subtyping
* **Consensus agglomerative clustering**: discovers patient subgroups from expression features, using bootstrap aggregation for robust subtype identification.
* **Feature extraction**: identifies the minimal gene set that distinguishes subgroups or predicts an outcome, via Orthogonal Matching Pursuit or elastic-net stability selection. Reports per-gene selection frequency alongside ROC-AUC, accuracy, and F1.
* **Complex heatmap**: clustered gene-panel heatmaps annotated with clinical variables, mutation status, or other categorical tracks.
### Expression distributions
* **Single-gene, multi-gene, and signature expression**: distributions across patient groups, for one gene, a panel, or a signature.
* **Expression variability**: average variability of genes across groups, identifying those with high inter-patient heterogeneity.
* **Correlation analysis**: pairwise Pearson or Spearman correlation between genes, proteins, or signatures, revealing co-expression and regulatory relationships.
* **Scatter/UMAP**: sample embeddings via UMAP or PCA, coloured by metadata or continuous expression.
Example use cases
* Volcano plot of genes differentially expressed between responders and non-responders
* Tumour vs. adjacent normal tissue comparison with DESeq2
* Differential expression between stages, adjusted for age and gender
* CD8A expression by tumour stage (violin plot)
* Clustered heatmap of immune checkpoint genes across patients
* IFN-gamma signature scores in IO-treated vs. untreated patients
* Kaplan-Meier curves for PD-L1 high vs. low expressors
* Forest plot of a multivariate prognostic model combining signatures and clinical variables
* Hallmark pathway activity across tumour subtypes with ssGSEA
* Expression-based patient subtypes with consensus clustering
* UMAP of patients coloured by histological subtype
## Whole exome sequencing (WES)
WES captures protein-coding regions, enabling identification of somatic mutations and copy number variations (CNVs).
* **Mutation frequency analysis**: per-position mutation frequency across a cohort, identifying hotspots within a gene. Supports variant allele frequency (VAF) display per patient, distinguishing clonal from subclonal mutations.
* **Oncoprint analysis**: a genes × samples alteration matrix showing mutations and CNVs across a gene panel, characterising the mutational landscape and revealing co-occurrence and mutual exclusivity.
* **Alteration co-occurrence**: co-occurrence of alterations (for example TP53 and ERBB2), stratified by clinical variables such as gender or stage.
* **Survival with genomic stratification**: mutation status against OS or PFS via Cox models or Kaplan-Meier curves, with multivariate adjustment for clinical covariates.
* **Expression-alteration integration**: expression heatmaps annotated by WES-derived CNV status, showing how genomic alterations track with transcriptional change.
Example use cases
* Oncoprint of TP53, KRAS, and EGFR alterations across NSCLC patients
* Lollipop plot of KRAS mutation hotspots (G12C, G12D)
* Survival curves comparing TP53-mutant and wild-type patients
* EGFR/TP53 co-occurrence by stage (stacked bars)
## Single-cell RNA sequencing
Single-cell RNA-seq resolves cellular heterogeneity, revealing cell states, rare populations, and intercellular communication.
* **Sparse aggregation**: percent-expressing cells and mean expression per group for a gene panel across cell types — the basis for single-cell dotplots.
* **Proportion analysis**: cell-type composition per sample, showing how populations vary across patient groups or conditions.
* **Cell-cell communication**: ligand-receptor signalling between cell types, inferred with LIANA and supporting CellChat, CellPhoneDB, and NATMI scoring.
* **Co-expression analysis**: Jaccard-index overlap between two genes or signatures, quantifying the fraction of cells expressing only A, only B, or both.
* **Sparse correlation**: pairwise Pearson or Spearman correlation between features, optionally stratified by cell type, to find co-regulated modules.
* **UMAP visualisation**: single-cell embeddings coloured by metadata or continuous expression.
* **Pseudobulk differential expression**: aggregates single-cell counts to patient-level profiles, then runs DESeq2 between clinical groups — across all cells or within one cell type.
* **InferCNV**: infers copy-number variation from single-cell expression, clustering CNV profiles into clones via Leiden clustering.
* **Single-cell enrichment (ssGSEA)**: pathway and signature activity scored per cell or per pseudobulk sample.
* **Consensus clustering**: patient subgroups from aggregated single-cell features.
Example use cases
* T-cell proportions in responders vs. non-responders (boxplot)
* Dotplot of checkpoint genes (PDCD1, CTLA4, LAG3) across immune populations
* Tumour-immune ligand-receptor crosstalk with CellPhoneDB scoring
* Malignant subclone identification with InferCNV
* IFN-gamma signature activity in CD8+ T cells with ssGSEA
* Pseudobulk differential expression in macrophages between tumour stages
## Spatial transcriptomics
Spatial transcriptomics preserves tissue architecture while measuring expression, enabling study of cellular niches and the tumour microenvironment in situ.
* **Spatial slide display**: Visium spots on tissue sections, coloured by metadata, gene expression, deconvolution results, or colocalization.
* **Spatial deconvolution**: cell-type composition estimated per spot, displayed as pie charts overlaid on the tissue.
* **Gene-gene colocalization**: spatial co-occurrence of two genes, identifying regions of coordinated expression.
* **Gene-cell type colocalization**: where a gene's expression spatially overlaps deconvolved cell-type abundance.
* **Spatial InferCNV**: copy-number variation inferred by comparing tumour against normal regions defined by spatial annotation.
* **Sparse aggregation, proportion, co-expression, and correlation**: as for single-cell, computed per spatial region or tissue compartment.
Example use cases
* Tumour microenvironment map: slide coloured by deconvolved immune composition
* CD8A expression across tumour and stroma regions
* Immune-tumour interface: colocalization of CXCL9 with T-cell abundance
* Cell-type proportions in tumour core vs. invasive margin
* PD-L1/PD-1 signalling at tumour-immune boundaries
* Spatially distinct malignant clones with spatial InferCNV
### Single-cell compared with spatial
| | Single-cell RNA-seq | Spatial transcriptomics |
| --------------- | ------------------------ | --------------------------------- |
| Resolution | Individual cells | \~1-10 cells per spot |
| Tissue context | Dissociated (lost) | Preserved architecture |
| Key insight | Cell identity and states | Spatial organisation and niches |
| Deconvolution | Not required | Required for cell-type inference |
| Unique analyses | Trajectory inference | Colocalization, region comparison |
## H\&E histomics
Histomics applies deep learning to whole-slide images, extracting quantitative features from standard pathology slides.
* **Cell-type density proportions**: proportions of AI-detected cell densities — tumour cells, lymphocytes, fibroblasts, macrophages — compared across clinical groups.
* **Aggregation analysis**: distributions of histomics features (cell densities, morphological metrics, spatial features) across groups.
* **Correlation analysis**: between histomics features, or between histomics and molecular features, linking morphological and molecular phenotypes.
* **Survival analysis**: Cox modelling and Kaplan-Meier curves stratified by histomics features such as high vs. low tumour-infiltrating lymphocyte (TIL) density.
* **Multivariate prognostic models**: histomics features combined with clinical variables and molecular markers, assessing independent prognostic value.
* **Complex heatmap**: clustered heatmaps of histomics features annotated by clinical variables or molecular subtypes.
* **Clustering and feature extraction**: morphologically-defined subtypes, and minimal interpretable feature panels.
### Feature categories
| Category | Examples |
| ---------------------- | -------------------------------------------------------------------- |
| Cell densities | Tumour cell, lymphocyte, fibroblast, macrophage density |
| Cell proportions | TIL proportion, stromal proportion, tumour proportion |
| Morphological features | Nuclear size, nuclear shape, pleomorphism |
| Spatial features | Cell clustering, spatial heterogeneity, tumour-stroma interface |
| Tissue architecture | Gland formation, necrosis fraction, inflammatory infiltrate patterns |
### Pathology Explorer
A dedicated module adds slide-level and tile-level capabilities: thumbnail display and an interactive pan-and-zoom viewer; filtering slides or tiles by any histomics feature; cohort feature statistics; AI model predictions overlaid on tiles; Kaplan-Meier analysis stratified by histomics feature; and export to Parquet.
Example use cases
* Tumour-infiltrating lymphocyte density by treatment response (boxplot)
* Cell-type composition across tumour stages (stacked bar)
* Kaplan-Meier curves for high vs. low lymphocyte density
* Cox model combining TIL density, CD8A expression, and stage
* Slides with the highest lymphocyte density, then the tiles with the highest immune infiltration
* Minimal histomics panel distinguishing responders, via feature extraction
* CD8+ T-cell density on H\&E correlated with CD8A bulk expression
## Clinical variables
Clinical data gives molecular findings their context, supporting survival modelling, cohort characterisation, and treatment response analysis.
* **Clinico-pathological characterization**: descriptive "Table 1" statistics — mean (SD) or median \[IQR] for continuous variables, n (%) for categorical — with between-group hypothesis testing.
* **Survival and Kaplan-Meier analysis**: Cox modelling and stratified curves for clinical variables such as stage, treatment, or biomarker status.
* **Timeline analysis**: per-patient treatment histories, including therapy sequences, durations, and responses.
* **Treatment flow diagrams**: patient flow across treatment sequences, such as 1L to 2L to 3L transitions.
* **Trajectory analysis**: longitudinal measurements over time, such as tumour size, PSA, or biomarkers.
* **Population overview**: patient count matrices by condition and dataset, for quick availability assessment.
* **Multi-variable and indication summaries**: value distributions of several variables per group, and per-indication summaries of the relevant features.
* **Consensus clustering and feature extraction**: patient subgroups and minimal actionable feature sets from clinical numerics.
Timeline, treatment flow, and trajectory analyses need treatment and measurement event data, which MOSAIC provides.
### Variable types
| Type | Examples | Typical analyses |
| ------------- | --------------------------------------------- | ------------------------------------------------- |
| Categorical | Gender, stage, grade, histology | Frequency tables, chi-square, stratified survival |
| Continuous | Age, BMI, tumour size, lab values | Mean/median comparison, t-tests, correlation |
| Time-to-event | OS, PFS, DFS | Kaplan-Meier, Cox regression, log-rank |
| Longitudinal | Treatment sequences, measurement trajectories | Timelines, flow diagrams, trajectory plots |
| Ordinal | ECOG status, tumour grade, line of therapy | Trend tests, stratified analyses |
Example use cases
* Table 1 comparing age, gender, and stage between responders and non-responders
* Kaplan-Meier curves stratified by tumour stage
* Multivariate Cox forest plot of age, stage, and grade effects on OS
* Per-patient therapy sequences as a Gantt chart
* 1L to 2L to 3L therapy transitions as a Sankey diagram
* PSA or tumour size tracked over the course of treatment
* Data completeness: summary table of missingness rates
## Multimodal integration
Combining modalities is where K-Pro's analytical range matters most, supporting cross-platform validation and robust biomarker development.
### Cross-modality concordance
Bulk RNA-seq can be compared against single-cell, spatial, or proteomics data; single-cell against spatial; and histomics against either expression or single-cell composition. These comparisons validate findings across platforms — for example confirming that a bulk expression signal localises to the expected tissue compartment, or that H\&E-derived TIL density agrees with CD8A expression.
### Integration analyses
* **Expression by mutation or CNV status**: transcriptional programs associated with an alteration, and dosage-sensitive genes.
* **Transcriptomics with histomics**: expression correlated against cell densities, expression stratified by histomics-defined groups, and joint clustering.
* **Proteomics with transcriptomics**: protein-RNA correlation, differential expression concordance, and multi-omics clustering.
* **Single-cell with spatial**: deconvolution validation, ligand-receptor analysis under spatial proximity constraints, and signature mapping onto spatial coordinates.
* **Clinical with molecular**: Table 1 across molecularly-defined groups, survival by molecular subtype, covariate-adjusted molecular effects, and treatment response by biomarker.
* **Advanced**: consensus clustering, feature extraction, and multivariate prognostic models over mixed features from several modalities at once.
Example use cases
* Patient clustering on bulk RNA signatures, histomics densities, and clinical numerics together
* Minimal gene + histomics + clinical panel per subtype, via feature extraction
* Protein-RNA concordance to validate a transcriptomic biomarker at protein level
* Multivariate survival model with age, stage, TP53 mutation, CD8A expression, and TIL density
* Pathway activity in KRAS-mutant vs. wild-type patients
* Genes with copy-number-driven expression
* Response biomarker discovery combining differential expression, mutation enrichment, cell-type composition, and histomics features
* Treatment timeline coloured by molecular subtype
## Statistical methods
| Method | Application |
| --------------------------- | ---------------------------------------------------------------------- |
| DESeq2 | Negative binomial model for count-based differential expression |
| Welch's t-test | Two-group comparison for normalized expression |
| Wilcoxon rank-sum | Non-parametric two-group comparison, for example of histomics features |
| Pearson / Spearman | Pairwise correlation between features |
| Cox proportional hazards | Multivariate survival modelling |
| Kaplan-Meier estimation | Survival probability curves |
| Log-rank test | Survival curve comparison between groups |
| Chi-square / Fisher's exact | Categorical variable comparison |
| ANOVA / Kruskal-Wallis | Multi-group comparison |
| ssGSEA | Per-sample pathway enrichment scoring |
| GSEA | Ranked list enrichment against gene sets |
| Consensus clustering | Robust unsupervised patient subtyping |
| Elastic-net / OMP | Sparse feature selection for biomarker discovery |
| LIANA | Ligand-receptor inference for cell-cell communication |
| InferCNV | Copy-number inference from expression data |
## Dataset coverage by modality
Which analyses you can run depends on the modalities present in your dataset. Availability varies by plan and by data access agreement — see **Browse the dataset catalog** and **What's included in your plan**.
| Dataset | Primary focus | Modalities |
| -------------------- | ------------------------- | -------------------------------------------------------- |
| MOSAIC | Multi-indication oncology | Bulk RNA, single-cell, spatial, WES, histomics, clinical |
| TCGA | Pan-cancer atlas | Bulk RNA, WES, histomics, clinical |
| CPTAC | Proteogenomics | Bulk RNA, proteomics, WES (variants), clinical |
| GTEx | Normal tissue | Bulk RNA, clinical |
| CellxGene Breast | Breast cancer | Single-cell, clinical |
| CellxGene HNSCC | Head and neck cancer | Single-cell, clinical |
| CellxGene NSCLC | Lung cancer | Single-cell, clinical |
| CellxGene Intestinal | GI tract | Single-cell, clinical |
MOSAIC offers the broadest multimodal coverage and is the only dataset supporting spatial transcriptomics.
---
# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.
## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation by asking a question.
Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters:
```
GET https://docs.owkin.com/explore-and-analyse-data/understand-your-data/analysis-methods.md?ask=&goal=
```
`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.
Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.