In cit-bioinfo/MS.liverK: Subtyping of Liver Cancer Samples According to Various Unsupervised or Supervised Classifiers Based on Gene Expression Data

Introduction

Cancer is a highly heterogeneous disease, with marked differences between patients. Even within a specific type of cancer, such as hepatocellular carcinoma (HCC), heterogeneity persists on several points: diverse genetic mutations, resulting in diverse cancer phenotypes. Therefore, characterizing these differences within a cancer type is highly important to better understand the mechanisms that undergo cancer maintenance and progression. A more thorough classification of cancer samples into different subtypes also has tremendous benefits in terms of precision medicine, as is allows clinicians to target therapies to the specific disease of each patient. A striking example of targeted therapy is the treatment of resistant patients by anti-PD1 antibody immunotherapy. To better understand cancer heterogeneity and identify patients more likely to respond to a specific treatment, the notion of molecular subgroups has emerged.

In HCC, several teams have designed unsupervised or supervised classification strategies. MS.liverK provides easy-to-use functions to characterize HCC samples from transcriptomic data. It implements 6 different molecular subtypes classification and scores 45 genes signatures about molecular subtypes, prognosis or biologial pathways.

Molecular classification provided in MS.liverK

Lee

In 2004, Lee and colleagues established a two-classes unsupervised classification of HCC (Lee et al., 2004). Their study was based on the gene expression profiles of 91 samples, which they divided into two classes highly associated to survival.

Boyault

Three years later, another team proposed a 6-groups unsupervised classification (Boyault et al., 2007). Their subgroups were found by unsupervised clustering of transcriptome data on 57 HCC samples and they have been linked to clinical and genomic information: G1 and G2 are linked to HBV virus’ DNA presence and differ by the main mutations found. G3 is characterized by TP53 mutation and overexpression of cell cycle genes. G5 and G6 have mutations that activate Wnt pathway. G4 is an heterogeneous group.

Chiang

Chiang and colleagues, in 2008, derived a classification of HCC into 5 subgroups from the analysis of 91 hepatitis C-related HCC samples (Chiang et al., 2008). Their classes are labeled “CTNNB1”, “Proliferation”, “Inflammation”, “Polysomy chr7” and an unannotated class. “CTNNB1” is enriched in mutations on the CTNNB1 gene, “Proliferation” class is enriched in IGF1 and RPS6, “Polysomy chr7” is defined by polysomy of chromosome 7.

Hoshida

In 2009, Hoshida and colleagues proposed a classification into 3 subgroups, from an analysis of genes expression profiles in nine cohorts, totalizing 603 patients. Their subgroups correlate with tumor size and cellular differenciation. Subgroup S1 contains samples with abnormal activation of Wnt pathway, S2 is characterized by proliferation and MYC and AKT activation, while S3 correlelates with hepatocyte differenciation.

Roessler

In 2010, Roessler and colleagues (Roessler et al., 2010) presented a two subgroup classification to identify the risk of metastasis-related recurrence. They used 2 cohorts, totalizing 386 patients.

Prognosis and biological pathway signatures

MS.liverK implements a prognostic signature published by Nault et al. (2013). Several teams have published gene-signatures related to biological functions/pathways involved in liver cancer oncogenesis. These gene-signatures relate to TGFB1 signalling (Coulouarn et al., 2008), MET pathway (Kaposi-Novak et al., 2006), stemness (Oishi et al., 2012; Yamashita et al., 2008), EPCAM (Yamashita et al., 2008) and hypoxia (van Malenstein et al., 2010). These signatures are implemented in MS.liverK.

Example of use

Code

To give an example of use of MS.liverK, we will apply it to a cohort of 91 patients from GSE20238 (Minguez et al., 2011) that is provided with the package.

First, load MS.liverK

library("MS.liverK")

To load the GSE20238 data, enter :

data("GSE20238")

This dataset consists of two data frames. The first one is called GEP (for Gene Expression Profiles). Is regroups the transcriptomic data from the 91 samples, at the probe set level. Data has previously been normalized and logged. The second data set, called AnnotProbeset, has two columns and one line per probe set. It contains the probe set's ID and the corresponding HUGO gene symbol.

We can now run the characterization of the samples, by using :

subtypes <- MS.liverK.subtypes(probesData=GSE20238$GEP,probesSymbols=GSE20238$AnnotProbeset)

subtypes now contains a list with 3 elements :

1. subtypes$molecularSubtype regroups all information concerning molecular subgroups classification :
2. subtypes$molecularSubtype$prediction contains the classes predicted by MS.liverK
3. subtypes$molecularSubtype$signatures contains the score of subtypes gene signatures
4. subtypes$prognosis regroups aresults for prognosis :
5. subtypes$prognosis$prediction contains the 2 classes prognosis output from Nault's signature (Nault et Zucman Rossi, 2013)
6. subtypes$prognosis$signatures is NULL by default, itcontains the score of prognosis gene signatures if the parameter PrognosticSignatures has been set to TRUE (advanced users only)
7. subtypes$biologicalPathwaysSignatures contains the scores of biological pathways signatures.

MS.liverK provides a function to discretize continuous signatures scores. To use it, run :

subtypes_discretized <- MS.liverK.discretize(subtypes)

This transforms the continuous scores into a discrete score from 1 to 5. This discrete score is data-dependant : the lower scores of the series will be set to 1 and the higher to 5.

To visualize the result, you can then run :

x11(width = 8.3, height = 11.7)
MS.liverK.plot(subtypes)

or

x11(width = 8.3, height = 11.7)
MS.liverK.plot(subtypes_discretized)

Dataset conversion

The predictors used by MS.liverK have been optimized for data aquired using Affmetrix Human Genome U133 Plus 2.0 Array. For data measured with other platforms, we encourage the users to convert it using the provided MS.liverK.convert function. To this end, the dataset must be already aggregated by gene, with HUGO gene symbols as row names, and samples in columns. For an optimal functionment, it is advised to have at least 50 samples. To convert your data set, simply replace data by the name of your dataset in the following code:

MS.liverK.convert(data)

cit-bioinfo/MS.liverK documentation built on Aug. 14, 2019, 10:34 a.m.