The Data Explained

The PeCan Knowledgebase consists of curated pediatric cancer genomics data, including variants, mutational signatures, gene expression data, and histological slide images from ~9000 hematological, CNS, and non-CNS solid tumor samples.

Variants

Oncoprint Methods

All oncoprint visualizations are generated using ProteinPaint’s mutational landscape study views.

Oncoprint Data

Variant data is sourced from published studies from St. Jude Children’s Research Hospital, NCI-TARGET TARGET, German Cancer Research Center DKFZ, and Shanghai Children’s Medical Hospital. A user can access associated studies by clicking Associated Study where applicable.

Tip
Gene lists are not curated. Access specific Associated Study links for subsets of data, such as the Medulloblastoma example here.

Variant Prevalence Methods

Variant Prevalence uses custom barplots to show the prevalence of each variant type. For some diagnoses, curated gene pathways are available, showing the top 20 genes; otherwise, the top 50 genes are displayed.

Mutation Class Glossary

Mutation Class	Description
MISSENSE	A substitution variant altering protein coding.
FRAMESHIFT	Insertion or deletion variant altering the protein coding frame.
NONSENSE	Premature stopgain or stoploss altering protein coding.
PROTEINDEL	Deletion removing codons without altering the protein coding frame.
PROTEININS	Insertion adding codons without altering the protein coding frame.
SPLICE	Variant near an exon edge potentially affecting splicing functionality.
SILENT	Substitution variant in the coding region that does not alter protein coding.
SPLICE_REGION	Variant in an intron within 10 nt of an exon boundary.
CNV LOSS	Copy number loss.
CNV GAIN	Copy number gain.

Warning
Silent mutations and UTR variants are excluded in some views. CNV classes are not displayed in ProteinPaint.

Variant Origin Glossary

Origin	Description
Germline	Variant found in a normal sample of a patient.
Somatic	Variant found only in a tumor sample.
Relapse	Variant found in a recurrence tumor.

Variant Prevalence Data

The following table outlines the datasets and variant types used in PeCan:

Dataset Label	Variant Type
BROAD	snv, indel, sv
DKFZ	snv, indel, sv
DKTK	snv, indel
MAGIC	snv, indel
PCGP	snv, indel, sv, fusions
SCMC	snv, indel, sv, fusions
SJCRH	snv, indel, sv
TARGET	snv, indel, cnv, sv, fusions
UTSMC	snv, indel

---

GenomePaint

GenomePaint Methods

Documentation for GenomePaint methods can be found here.

GenomePaint Data

The Pediatric2 dataset expands the original Pediatric dataset by adding non-coding variants and expression data. The additional data includes:

• Genome for Kids (PMID: 34301788)
• Shanghai Children’s Medical Center relapsed ALL cohort (PMID: 32632335)
• St. Jude’s Clinical cancer genomic profiling by three-platform sequencing study (PMID: 30262806)
• St. Jude’s Pan-neuroblastoma analysis data study (PMID: 33056981)

Reference Genome	Datasets
hg19	Pediatric, PAN-ALL
hg38	Pediatric, SJLIFE

Table 1. GenomePaint datasets. GenomePaint supports datasets for different reference genomes (hg19/hg38).

---

ProteinPaint

ProteinPaint Methods

Documentation for ProteinPaint methods can be found here.

ProteinPaint Data

The Pediatric dataset consists of somatic variants and tumor RNA-seq data, using the GRCh37/hg19 genome, as well as data from:

• Pediatric Cancer Genome Project (PCGP)
• NCI TARGET cohort
• Pan-TARGET study (PMID: 29489755)
• Shanghai Children’s Medical Center T-ALL cohort (PMID: 32632335)

The GRCh38/hg38 dataset was created through liftover of genomic variants and gene internals. Additional data is drawn from NCI’s Genomic Data Commons (GDC) and ClinVar, which is supported by NIH, NHGRI, and NCI.

Reference Genome	Datasets
hg19	Pediatric, COSMIC, ClinVar
hg38	Pediatric, GDC, COSMIC, ClinVar

Table 2. ProteinPaint datasets. ProteinPaint supports datasets for different reference genomes (hg19/hg38).

---

Mutational Signatures

Methods

COSMIC SBS signatures (v3.3) were identified using SigProfilerExtractor (v1.1.20) from pediatric tumor samples grouped by subtype.

Data

Sample data from G4K, PCGP, Clinical Pilot, and Real-Time Clinical Genomics (RTCG) were used for mutational signature detection.

Tip
Example data from 790 tumor samples can be found in Figure 5 of McLeod et al.

---

Expression

t-SNE Methods

RAPID RNA-Seq workflow was used to generate RNA-seq expression count data and visualized via ProteinPaint’s scatterplot.

t-SNE Data

Sample data comes from G4K, PCGP, Clinical Pilot, and Real-Time Clinical Genomics studies.

Tip
Example data from 1,574 RNA-Seq samples can be accessed here.

Gene Expression Methods

DESeq2’s Median of Ratios^1 was applied to ~5500 tumor samples for gene expression analysis. We advocate for this methodology over TPM/CPM due to its stability and resistance to outliers.

Key Benefits:

1. Assumes most genes are not differentially expressed, providing a stable reference.
2. Robust against highly expressed genes dominating total read counts.
3. More reliable differential expression results due to geometric mean normalization.
4. Consistent and comparable expression values across samples.
5. Love, M. I., Huber, W., & Anders, S. (2014). Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biology, 15(12), 550. https://doi.org/10.1186/s13059-014-0550-8

---

Histology

Methods

H&E-stained histology slides in .svs format are uploaded and processed in Azure. Currently, each sample has one slide, with plans to support multiple slides per sample.

Data

Histological images are provided by the COMET Blue Sky Initiative, led by Dr. Mike Dyer. A large subset of COMET slides are pending publication.

Interested in collaborating?
Reach out to us at support@stjude.cloud.