ELIXIR Single Cell Omics Benchmark FAIRification

Project Overview

In the initial phase, this project implements comprehensive FAIRification of the BE1 benchmark dataset from the ELIXIR Single Cell Omics Implementation Study (WP3), transforming the original 10XGenomics scRNA-seq experiment into a FAIR-compliant, machine learning-ready resource with controlled cancer heterogeneity using seven lung cancer cell lines with distinct driver mutations.

29,528

Total Cells Sequenced

7

Cancer Cell Lines

8

Samples (+PBMC)

80.0%

FAIR Score

Controlled Heterogeneity

Seven lung cancer cell lines with distinct driver mutations: EGFR Del19, KRAS G12S, MET Del14, BRAF G469A, ERBB2 V842I, ROS1 fusion, ALK fusion, plus PBMC control.

PC9: EGFR Del19 (4,492 cells)
A549: KRAS G12S (6,898 cells)
NCI-H596: MET Del14 (2,965 cells)
NCI-H1395: BRAF G469A (2,673 cells)
DV90: ERBB2 V842I (2,998 cells)
HCC78: ROS1 fusion (2,748 cells)
CCL.185-IG: ALK fusion (6,354 cells)
PBMC: healthy donor control

View Cell Line Details

Single Cell Schema v0.4

Complete FAIRification using Darwin Core extension for single-cell RNA-seq with 16 standardized metadata sheets covering all experimental aspects.

Study, person, sample metadata
Dissociation, cell suspension protocols
Library prep & sequencing parameters
Raw & downstream processing
Expression data files & settings

View Schema Details

ML-Ready Benchmark

Designed for developing and validating computational methods for cancer heterogeneity analysis with controlled ground truth.

R Shiny App for controlled mixtures
BE1-500 validation dataset
CCLE bulk transcriptome comparison
Cell Ranger v7.1.0 processed
intron-mode quantification

Shiny App

FAIR Assessment

Comprehensive FAIR evaluation of the BE1 dataset after transformation:

80

Findable

Persistent IDs: GEO, DOI, SRA, BioProject, figshare

80

Accessible

FTP, SRA Run Selector, CC BY-NC 4.0 licence

75

Interoperable

Single Cell Schema, controlled vocabularies, MTX/HDF5 formats

85

Reusable

Full provenance, QC metrics, validation against CCLE

FAIRification Workflow

Systematic transformation of the BE1 dataset following Single Cell Schema standards:

1

Data Collection

7 lung cancer cell lines + PBMC, 29,528 cells sequenced on 10X Genomics

2

Schema Mapping

Single Cell Schema v0.4 with 16 metadata sheets

3

Identifier Enrichment

GEO, SRA, figshare DOIs, PMID, BioProject

4

QC Validation

rCASC mito/ribo filtering, CCLE comparison, BE1-500 clustering

5

ML Enablement

R Shiny mixing app, feature extraction, benchmark datasets

Per-Cell Line Statistics

Cell Ranger v7.1.0 statistics for each cell line (from publication):

Cell Line	Driver Mutation	Cells	Mean Reads/Cell	Median Genes	Sequencing Saturation
PC9	EGFR Del19	4,492	28,984	2,868	47.8%
A549	KRAS G12S	6,898	29,864	3,307	60.1%
CCL-185-IG	EML4-ALK fusion	6,354	29,859	2,810	60.6%
NCI-H1395	BRAF G469A	2,673	27,972	2,822	48.0%
DV90	ERBB2 V842I	2,998	26,756	2,387	41.4%
HCC78	ROS1 fusion	2,748	24,700	2,570	38.7%
NCI-H596	MET Del14	2,965	28,666	3,113	56.7%
PBMC	Healthy control	~500	-	-	-

Total sequenced: 2.46 billion reads, ≥71.26% bases ≥Q30 on NovaSeq X Plus

Key Features & Innovations

Single Cell Schema v0.4

Complete Darwin Core extension implementation with 16 standardized metadata sheets covering all experimental aspects from dissociation to expression data.

Controlled Heterogeneity

Seven distinct driver mutations with partially overlapping pathways: EGFR, ALK, MET, ERBB2, KRAS, BRAF, ROS1 - ideal for benchmarking.

R Shiny Mixing App

Interactive tool to generate controlled mixtures of cell lines at user-defined ratios for creating custom benchmark datasets.

Comprehensive Validation

BE1-500 clustering validation, CCLE bulk transcriptome comparison, external dataset integration (Tian et al., Aissa et al.)

Validation Results

BE1-500 Clustering

Louvain resolution 0.1: 6 clusters

Cluster composition: Each cluster predominantly from specific cell line

Note: A549 and CCL-185-IG co-cluster (syngeneic)

CCLE Comparison

Method: Hierarchical clustering of pseudo-bulks

Result: Strong alignment between single-cell pseudo-bulks and CCLE bulk transcriptomes

QC Filtering

Tool: rCASC mitoRiboUmi plot

NCI-H1395: Reduced from 2,673 to 1,939 cells after QC

Thresholds: mito <20%, genes >200, UMI >500

Data Access & Resources

Primary Data

GEO: GSE243665 (count matrices)
SRA: PRJNA1019356 (FASTQ)
BioProject: PRJNA1019356
figshare: 10.6084/m9.figshare.23939481.v1

Supplementary Data

CCLE data: 10.6084/m9.figshare.23284748.v1
Driver gene targets: IPA & TRRUST relations
Code & docs: 10.6084/m9.figshare.24744996.v1
multi_gex.csv: Cell Ranger config in GEO

Tools

Shiny App: https://github.com/kendomaniac/BE1
Docker: repbioinfo/cellranger.2023.7.1.0
Cell Ranger: v7.1.0 (intron mode)
Reference: GRCh38-2020-A

Quick Start Guide

1. Download Data

# GEO count matrices (MTX format)
wget -r ftp://ftp.ncbi.nlm.nih.gov/geo/series/GSE243nnn/GSE243665/suppl/

# Or via figshare
wget https://figshare.com/ndownloader/files/43908748 -O BE1_matrices.zip

2. Reproduce Cell Ranger

docker run -v /data:/data \
  repbioinfo/cellranger.2023.7.1.0 \
  cellranger multi --id=BE1run1 \
  --csv=/data/multi_gex.csv

3. Generate Controlled Mixtures

# Access Shiny App
http://aisc.hpc4ai.unito.it:3838/

# Select cell lines and ratios
# Download custom sparse matrix
# Cell barcodes contain cell line names

Citation and License

Dataset Reference Publication

Arigoni M, Ratto ML, Riccardo F, Balmas E et al. (2024). A single cell RNAseq benchmark experiment embedding "controlled" cancer heterogeneity. Scientific Data, 11(1), 159. https://doi.org/10.1038/s41597-024-03002-y

License

This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.