Frequently Asked Questions

Why is it called COHORTome?

The name combines cohort, the people or samples studied together, with the scientific suffix -ome, which suggests a connected view across clinical, assay, imaging-derived, and analytical data.

The typography makes that meaning explicit: COHORT is the primary term and ome is a quieter suffix. A secondary reading, "cohort to me", reflects the idea of bringing an authorized cohort into a researcher's workspace, but it is wordplay rather than the main identity.

What is ClarID?

ClarID is an identifier format that encodes selected biomedical metadata in a human-readable code, with a compact form called a stub. It supports identifiers for subjects and biosamples.

In COHORTome, ClarID is an optional annotation alongside the original identifier. Click a ClarID label to inspect its decoded fields, including the condition name when available. You can also search for a patient by ClarID.

A Subject ClarID shown beside a sample describes the linked patient; it is not a biosample ClarID. The original patient and sample IDs, together with their source dataset, remain the identifiers COHORTome uses to link records.

ClarID is not anonymization

A code can reveal metadata such as a condition. Treat it with the same care as the record it describes; a compact stub is not encryption.

See the ClarID paper and clarid-tools.

Is COHORTome only for 3TR?

No. COHORTome is generic open-source software that can be deployed for other approved biomedical datasets. It is being developed in the context of 3TR, but does not depend on a particular consortium or study.

Read Project context for that project context.

Is COHORTome a data warehouse?

No. Warehouses and institutional services remain responsible for storing and distributing source data. COHORTome receives authorized releases and provides a workspace for linking records, exploring patients and assays, finding cohorts, and running traceable analyses.

Read Scope And Boundaries for the division of responsibilities.

Does COHORTome process raw medical images?

No. COHORTome can open an authorized DICOM study in OHIF, but it does not implement a viewer, PACS, segmentation pipeline, or pixel-level model. Qualified imaging systems and teams produce processed measurements; COHORTome can combine those derived features with clinical and assay data.

Are the bundled MRI images realistic multiple-sclerosis scans?

No. Their metadata exercises an MS brain-MRI workflow, but the visible pixels are generated procedurally without patient images. They are smoke-test fixtures for DICOM linkage and OHIF rendering, not medically realistic MS images or scientific evidence.

Is a language model required?

No. Supported questions and analyses can run through deterministic application logic. An administrator may optionally configure approved language-model providers for more flexible query planning and result explanation. Dataset policy still determines what information, if any, may be sent.

Read Governed AI for the operating model.

How does COHORTome take RSEM expected counts into limma-voom?

RSEM expected counts estimate the fragments assigned to each gene. They remain non-negative count-scale measurements, but can be fractional because RSEM distributes ambiguously mapped fragments across possible transcripts or genes. They are therefore not the same input as integer raw read counts. COHORTome routes assays declared as rsem_expected_counts to a dedicated limma-voom analysis; its DESeq2 catalog entry is reserved for assays declared as raw_counts.

The current limma-voom runner performs these steps:

1. It aligns matrix columns with the selected samples, checks for finite, non-negative values, requires one sample per patient, and requires at least two samples in both Cohort A and Cohort B.
2. It creates an edgeR DGEList, removes weakly expressed genes with filterByExpr(..., group = cohort), and calculates TMM normalization factors with calcNormFactors.
3. voom transforms the retained values to log2 counts per million and estimates the mean-variance trend and observation-level precision weights.
4. limma fits the weighted linear model ~ cohort. Cohort B is the reference, so the tested cohort coefficient and reported log2 fold change are Cohort A minus Cohort B. A positive value means higher expression in Cohort A.
5. eBayes applies limma's empirical-Bayes variance moderation. topTable reports every retained gene, with Benjamini-Hochberg adjusted p-values; the run summary counts genes with adjusted p-value below 0.05.

Exploratory analysis

The current model is deliberately simple: it contains the cohort term only. It does not yet adjust for age, sex, recruitment center, treatment, batch, RIN, or other technical or clinical covariates. Treat it as an exploratory result until an appropriate covariate-aware design is specified and validated for the study.

The short sentence shown in the Analysis thread is only the run synopsis. The completed job's folded Result files section retains the full gene-level table (de_results.tsv), volcano-point table, volcano plot, machine-readable summary, run parameters, provenance bundle, and command logs.

Is COHORTome intended for clinical decisions?

No. It is an exploratory research workbench. Results require appropriate scientific review and must not be presented as diagnosis, treatment advice, or validated clinical decision support.