Look one variant up in five places
and five answers come back
One KRAS G12C goes by a coordinate, an HGVS string, an rs number and a COSMIC id β and its hg19 and hg38 positions differ by 152,935 bases.
Yet most of the actual work is joining: attaching recurrence, clinical calls, population frequency and actionability to your own variant table. When that join slips, it throws no error β conclusions appear that were never there, and real ones disappear.
Over these 20 lessons you perform those joins by hand on a real 982-row MAF and 2,731 samples from four cBioPortal studies, and count exactly where they leak. Python runs right inside this page.
How you'll learn
You join the same data on four different keys and count how many rows go missing or spurious each time. Nothing raises an error.
Within 982 rows, 17 protein names already point at several coordinates, and the EGFR exon-19 deletion sits there under two of them. This is not a hypothetical.
The original TCGA MAF, 2,731 samples across four cBioPortal studies, and five real gene panels. Not a stand-in dataset.
COSMIC, ClinVar, gnomAD and OncoKB records are absent β and the final lesson's card prints those four blanks rather than filling them in.