GDC-GRIN2 tests GDC copy number alterations and mutations with the Genomic Random Interval (GRIN) statistical model (Pounds, S. et al. Bioinformatics 2013). Pick a cohort and the form opens with that GDC filter applied; set the lesion filters (consequences, CNV cutoffs, artifact masks) and click Run for a genome-wide Manhattan plot of per-gene significance and a ranked gene table.
The cohorts show different shapes of result: a copy-number-driven genome, a CNV-quiet one, a hypermutated one, and cohorts built by diagnosis, disease type or primary site rather than by project. We checked case counts against the GDC API in August 2026 and ran six of the nine at default settings; those quote what came back.
Loading…
One dot per gene, positioned by genomic coordinate and colored by lesion type. Height is the GRIN significance of that gene carrying more lesions than random intervals of the same size and count would, so a wide gene needs more hits than a small one to rise.
SNV/indel, CNV gain and CNV loss are tested separately, so a gene can be significant on one and
flat on the others. In TCGA-PAAD, CDKN2A clears both: mutated in 30 cases and
deleted in 36.
Every gene inside an amplified or deleted segment scores, so a driver arrives with its
neighbors. In TCGA-GBM the chr7 amplicon puts EGFR-AS1, ELDR and
SEC61G alongside EGFR at the top of the table.
CNV segments longer than the Max Segment Length box (2 Mb by default) are dropped, which excludes whole-arm and whole-chromosome events. In uveal melanoma, raise it and chromosome 3 loss appears where there was nothing.
The server pulls mutation and CNV data per case: 80 cases finish in seconds, the largest cohorts here take minutes. Repeat runs at the same settings come from cache, so start small when demoing.
The cap is on total lesions loaded (up to 250,000, scaled to the server's free memory), not on cases. TCGA-SKCM alone loads 222,588. A run that hits the cap says so, with how many samples were left out.
A GDC cohort selects cases, not data. Without a logged-in GDC session only open-access lesions are readable, and several large projects publish little or none. The run summary reconciles this: total samples, samples processed, samples carrying a lesion, samples in controlled-access projects, and samples dropped by the hypermutator cutoff. The projects most likely to thin out a cohort selected by diagnosis, disease type or primary site:
| Project | Cases | Open SNV/indel files | Controlled | Gene-level CNV cases |
|---|---|---|---|---|
| FM-AD (targeted panel, in many diagnosis and site cohorts) | 18,004 | 0 | 54,012 (all) | 0 |
| TARGET-AML | 2,492 | 806 | 163 | 73 |
| BEATAML1.0-COHORT | 826 | 501 | 353 | 0 |
| TARGET-WT | 652 | 42 | 644 | 0 |