Multi-omics deconvolution framework to profile the tumor immune microenvironment (TME) and predict immunotherapy response, developed at the Cancer Research Center of Toulouse (CRCT).
Key Contributions#
- Pipeline Development: Led the design and implementation of two containerized Snakemake pipelines — one for gene expression deconvolution, one for DNA methylation deconvolution — processing multi-omics data across multiple cancer cohorts.
- Novel Signatures: Contributed to deriving novel deconvolution signatures (BPRNACan, BPRNACan3DProMet, BPmetCan) for B cells, CD4/CD8 T cells, M0/M1/M2 macrophages, monocytes, neutrophils, NK cells, and cancer cells.
- Scale: ~7,000+ TCGA RNA-seq samples for signature derivation; immunotherapy response prediction validated on 166 patients across 6 external cohorts (melanoma, bladder, GBM, lung).
- 3D Chromatin Integration: Incorporated PCHi-C 3D chromatin contacts from GARDEN-NET into the BPRNACan3DProMet signature, bridging chromatin architecture and immune deconvolution.
Technical Approach#
Gene expression deconvolution (bulk RNA-seq: BCL → FASTQ → TPM via STAR/RSEM/Salmon) and DNA methylation arrays (Illumina 450K and EPIC; IDAT → beta values via ChAMP/minfi) are processed in parallel pipelines. Immunotherapy response prediction uses ElasticNet penalized logistic regression with leave-one-dataset-out (LODO) cross-validation across the 6 immunotherapy cohorts.
Technologies: Python · Snakemake · R · Docker · Singularity · STAR · RSEM · Salmon · Scikit-Learn · ChAMP · minfi · immunedeconv · EpiDish · CIBERSORTx · Tidyverse.