Gradient Biotech

Subjects and cohorts

Cohort intelligence connects datasets to participant metadata and enables group-level network comparison.

Subjects table

Subjects appear on the study home page when imported via CSV.

CSV import

Upload a subjects CSV with at minimum:

external_id,age,sex,group_label,diagnosis
S001,34,F,control,healthy
S002,38,M,patient,MCI

Extra columns are stored as flexible metadata for filtering and report context.

Link datasets

Assign datasets to subjects via the metadata API or group labels on individual datasets from the cohort page.

Group assignment

Navigate to Cohort analysis (/areas/neurology/studies/{id}/cohort).

For connectivity matrix datasets, assign a group label to each dataset directly in the assignment table. You need at least two groups with multiple datasets each for NBS.

Batch graph metrics

When group labels are assigned, run Batch graph metrics to compute BCT measures across all matrix datasets in the study. Results feed cohort metric comparison and NBS.

Network-Based Statistics (NBS)

Run NBS compares connectivity matrices between groups using permutation testing:

  • Significant connected components highlighted on an edge heatmap
  • Component size and intensity statistics
  • Exportable NBS report artifact

Requires completed uploads for both groups.

Cohort metric comparison

Compare cohort metrics summarizes group differences in global graph measures (clustering, efficiency, modularity, etc.) with effect sizes and statistical tests.

Predictive biomarkers (Prediction tab)

The cohort page's Prediction tab reuses the same group labels (or a per-dataset continuous target) to train a cross-validated classifier or regressor on connectivity edges, with leakage-free feature selection and permutation testing. See Predictive biomarkers for the full workflow.

Interpret cohort results

Use Interpret on the cohort page after NBS or cohort metrics complete. The narrative cites computed statistics — it does not invent values.

Typical cohort workflow

  1. Create study and upload connectivity matrices for all subjects
  2. Assign group labels (control vs patient, or treatment arms)
  3. Run batch graph metrics
  4. Run NBS for edge-level group differences
  5. Compare global metrics across groups
  6. Interpret the results, and check the runs page for the full run history