docs/presentation.md

In priesgo/TCGAome: MultiOmics (just RNAseq and RPPA...) Data Analyzer on TCGA data

TCGA MultiOmics Data Analysis

author: Pablo Riesgo ([e-mail]) date: 08/11/2015

The dataset

Retrieval

• TCGA normalized level 3 datasets for RNASeq and RPPA
• For tumor types breast cancer and ovary cancer

library("RTCGAToolbox")
firehose_datasets = getFirehoseDatasets()
firehose_dates= getFirehoseRunningDates()
BRCA.dataset = getFirehoseData (dataset="BRCA", runDate=firehose_dates[1], RNAseq2_Gene_Norm = TRUE, RPPA=TRUE)
OV.dataset = getFirehoseData (dataset="OV", runDate=firehose_dates[1], RNAseq2_Gene_Norm = TRUE, RPPA=TRUE)

The dataset

Preprocessing

• Removed samples not having data for RNASeq and RPPA
• Removed variables for genes not matching with HGNC gene names
• Removed RPPA variables for antibodies matching the same gene, only the antibody with the maximum absolute average expression was conserved
• Scaling
• Removed variables with zero or near zero variance
• Removed variables with missing values
• Removed variables with correlation over 0.7

The dataset

Samples

407 BRCA + 201 ovary cancer samples

Variables

12880 gene expression + 55 protein expression variables are far too many to plot...

Principal component Analysis

Gene expression

Protein expression

Principal component Analysis

Joint analysis

Hierarchical clustering

Gene expression

Protein expression

Hierarchical clustering

Joint analysis

Multiple Co-Inertia Analysis (MCIA)

The 2 first PCs explain 34.05% and 11.55% of the variance

Labeled items are the top 100 variables

sparse Least Square Regression (sPLS)

Samples

Variables

sparse Least Square Regression (sPLS)

Heatmap

Network

Results

Matches between MCIA and sPLS top 100 and known oncogenes and tumor supressor genes

Matches between MCIA and sPLS enriched pathways in Reactome

Other methods to study

• Canonical Correlation Analysis
• Regularized Generalized CCA (error with dataset)
• Regularized CCA (long computation)
• CCA with Elastic Net (did not found implementation)
• Procrustes (pending)

priesgo/TCGAome documentation built on May 25, 2019, 11:26 a.m.