Prepare your data
Choose the route that matches your question before changing the input table.
| What you have | Route | Required inputs |
|---|---|---|
| Expression and clinical data | Pathway-first | Expression, clinical table and gene sets |
| Named genes, signatures or scores | Exposure-first | Feature table and explicit feature names |
| A saved RDS, workbook or delivery folder | Read or plot | The saved result and the requested task |
Expression and clinical data
Use a numeric expression matrix with genes in rows and samples in columns, unless you explicitly set another orientation. The clinical table has one row per analysis subject and a unique sample ID column. Match IDs explicitly; column position alone is not a sample map.
Keep the original expression scale and any transformation recorded. TPM, log-transformed abundance and standardized scores are different inputs. Gene identifiers must match the gene sets; inspect coverage before interpreting scores. Bundled IOBR signatures, supported curated sets and user-supplied membership are described in the pathway-first chapter.
A feature table
Each row is one subject. Keep identifiers, features, endpoints and baseline covariates in separate named columns. For example:
| ID | Selected_score | Age | Sex | Response | OS_time | OS_status |
|---|---|---|---|---|---|---|
| S001 | 0.42 | 61 | F | 1 | 12.5 | 1 |
| S002 | -0.15 | 68 | M | 0 | 20.0 | 0 |
These two rows illustrate the format only. They cannot support a fitted analysis. Name the requested features explicitly so unrelated numeric columns stay out.
Define endpoints and covariates
- Response: record exactly what 1 and 0 mean, including how CR/PR and other categories are handled. Missing response is not automatically non-response.
- OS: provide time, event status, the event code and time unit. Check event and censoring counts after matching subjects.
- Covariates: use the agreed baseline adjustment set. A tutorial list such as Age, Sex and ECOG is an example, not a verified set for every study.
Check duplicate IDs, missing values, invalid categories and unsupported exposure contrasts before execution. Keep the full requested feature family for FDR; selecting a display subset does not redefine that family.