Detection of non-coding RNAs and assessment of their predictive value for mono therapies

Julia Nguyen^1,\$, Anthony Mammoliti^1,2,\$, Sisira Kadambat Nair¹, Emily So^1,2, Farnoosh Abbas Aghababazadeh¹, Christoper Eeles¹, Ian Smith^1,2, Petr Smirnov^1,2, Housheng Hansen He^1,2, Ming-Sound Tsao^1,2, Benjamin Haibe-Kains^{1,2,3,4,5,6,#}

¹Princess Margaret Cancer Centre, University Health Network, Toronto, Ontario, Canada\ ²Department of Medical Biophysics, University of Toronto, Toronto, Ontario, Canada\ ³Department of Computer Science, University of Toronto, Toronto, Ontario, Canada\ ⁴Ontario Institute of Cancer Research, Toronto, Ontario, Canada\ ⁵Vector Institute for Artificial Intelligence, Toronto, ON M5G 1L7, Canada\ ⁶Biostatistics Division, Dalla Lana School of Public Health, Toronto, ON M5T 3M7, Canada

^$ These authors contributed equally to the present work\ ^# Corresponding author: Benjamin Haibe-Kains, Princess Margaret Cancer Centre, University Health Network, Toronto, Ontario M5G 2C4 Canada

Transcript Quantification

Description: circRNA, mRNA, and corresponding gene expression was quantified across 48 cell line biological replicates from three pharmacogenomic datasets (gCSI, CCLE, GDSC2) with poly(A)-selected RNA-seq data. Consistency of transcript quantification was compared using UMAP. Transcript stability was assessed across the dataset pairs.

1. Reproducible Snakemake pipelines for running CIRI2 and CIRCexplorer2 on cell line biological replicates and lung cancer patient data can be found under: "/data/snakemake_pipelines"

2. Transcript expression quantification (run script: "/code/transcript_quantification.R")

3. Dimensionality reduction analysis of transcript quantification (run script: "/code/umap.R")

4. Computation of stability indices of transcripts across dataset pairs (run script: "/code/si_distribution.R")

circRNA

Description: CIRI2 and CIRCexplorer2 was ran on 48 cell line biological replicates (gCSI, CCLE, GDSC2) with poly-A selected RNA-seq data and lung cancer patient data with both poly-A + Ribo-depleted RNA-seq data. circRNA detection levels were compared between cell line and patient data. A survival analysis (Overall Survival) was performed using the Ribo-depleted lung patient data due to yielding high circRNA expression, in comparison to the cell lines.

1. Cell line biological replicate analysis - data under "/data/processedcellline" (run script: "/code/circRNAcelllines.R")

2. Lung adenocarcinoma patient analysis - data under "/data/processedlung" (run script "/code/circRNAlung.R")

3. Quantification of unique circRNA transcripts and distribution across RNA-seq selection protocols (run script "/code/uniquelungtranscripts.R")

4. Survival analysis on lung adenocarcinoma patient data (run script: "/code/survival_lung.R")

mRNA

Description: Features that may predict mRNA stability were processed, and influence was computed through supervised learning. The association between mRNA stability and drug response was investigated.

1. Generation of data.frame with transcript stability indices and genomic feature data - data under - "/data/transcriptstability" (run script: "/code/transcriptstability.R")

2. Identification of quantile usage for stable/unstable transcripts using Kuncheva Index (run script: "/code/overlap.R)

3. Feature influence after permuting each feature (n=20,000) (run script: "/code/feature_influence.R")

4. Assessing association between stability and predictive value for known gene biomarkers (run script: "/code/biomarker_forest.R")

5. Assessing association between stability and predictive value for across all genes (run script: "/code/histpredvalue.R")