Example data

The Polaris example project, polaris-epigenomics, comes with a cancer genomics database you can query right away. It joins three public datasets from the AWS Open Data Registry:

• TCGA (The Cancer Genome Atlas), from the NCI Genomic Data Commons, Data Release 46.0.
• Open Targets Platform, release 26.09.
• IDC (NCI Imaging Data Commons), version 25.

The data lives in one DuckDB database, epigenomics. Polaris opens it read-only.

The cohort

The cohort has 755 patients from two TCGA projects:

Project	Cancer type	Patients
TCGA-KIRC	Kidney renal clear cell carcinoma	363
TCGA-BLCA	Bladder urothelial carcinoma	392

Every patient in the cohort has clinical data and one primary tumour sample with RNA-seq expression, somatic mutations, copy number and DNA methylation. Some patients also have a matched normal tissue sample. These two projects were chosen because they have many patients with every data type, including radiology images in IDC.

Tables

TCGA

Table	Rows	What it holds
tcga_patients	755	One row per patient: clinical fields such as sex, age, stage and survival, plus imaging flags.
tcga_samples	1,030	One primary tumour sample per patient, and a matched normal sample where one exists.
tcga_files	3,334	The GDC open-access file behind each sample-level row.
tcga_genes	19,944	Protein-coding genes (GENCODE v36). gene_id is the Ensembl id without a version.
tcga_gene_expression	16.7 million	STAR gene counts, one row per sample and gene.
tcga_gene_methylation	17.4 million	Promoter methylation beta values per sample and gene.
tcga_gene_copy_number	14.1 million	Gene-level copy number. 2 is normal (diploid), 0 is a homozygous deletion.
tcga_somatic_mutations	131,062	Somatic mutations from whole-exome sequencing. Filter on is_nonsynonymous for protein-altering mutations.
tcga_project_disease_map	5	Links each TCGA project to Open Targets disease ids.

Open Targets

Table	Rows	What it holds
ot_targets	20,107	Protein-coding drug targets, with prioritisation flags.
ot_target_tractability	59,587	Tractability assessments by modality: small molecule, antibody, PROTAC and other.
ot_associations	32,717	Target to disease association scores, overall and by data type.
ot_known_drugs	3,634	Drugs and clinical candidates, one row per drug, disease and target.
ot_diseases	57	The diseases in scope and their sub-types.

IDC

Table	Rows	What it holds
idc_series	7,910	Metadata for imaging series: radiology (CT, MR, PET and X-ray), slide microscopy, and derived series such as segmentations. Images themselves are not included.

Provenance

data_sources lists each source with its version and URL. Ask Polaris “Where does this data come from?” to see it.

How the tables join

• TCGA tables join on patient_id and sample_id.
• tcga_genes.gene_id joins ot_targets.target_id, which connects mutations and expression to drug targets.
• tcga_project_disease_map connects a TCGA project to Open Targets diseases.
• IDC patient ids are TCGA case barcodes, so idc_series joins the cohort by patient.

You do not need to remember any of this. The project’s RULES.md and table docs describe the joins and the common mistakes, and Polaris reads them before it writes SQL.

Optional COVID-19 data

The example project can also connect to a Snowflake database, covid19, with public COVID-19 epidemiological data (Johns Hopkins, WHO, CDC and Google mobility tables). It is a separate topic from the genomics data. If it is not connected on your deployment, Polaris tells you so.

Questions to try

• How many patients in each cohort have a VHL mutation, and how many also have a PBRM1 mutation?
• Which genes are most often hypermethylated in KIRC tumours?
• Which targets associated with bladder cancer have an approved drug?
• What mix of imaging modalities does IDC have for our cohort?
• Do KIRC patients with a BAP1 mutation have shorter overall survival?