Chapter 3: cancer multi-omics paper

content/30.header.md

@@ -0,0 +1,9 @@
+## Chapter 3: Widespread redundancy in -omics profiles of cancer mutation states
+
+This chapter has been published in _Genome Biology_ (https://doi.org/10.1186/s13059-022-02705-y) as "Widespread redundancy in -omics profiles of cancer mutation states".
+
+**Contributions:**
+JC: conceptualization, methodology, software, visualization, writing - original draft, writing - review and editing
+BCC: methodology, writing - review and editing
+MC: methodology, writing - review and editing
+CSG: conceptualization, funding acquisition, methodology, supervision, writing - review and editing

content/31.abstract.md

@@ -0,0 +1,19 @@
+### Abstract
+
+#### Background
+
+In studies of cellular function in cancer, researchers are increasingly able to choose from many -omics assays as functional readouts.
+Choosing the correct readout for a given study can be difficult, and which layer of cellular function is most suitable to capture the relevant signal remains unclear.
+
+#### Results
+
+We consider prediction of cancer mutation status (presence or absence) from functional -omics data as a representative problem that presents an opportunity to quantify and compare the ability of different -omics readouts to capture signals of dysregulation in cancer.
+From the TCGA Pan-Cancer Atlas that contains genetic alteration data, we focus on RNA sequencing, DNA methylation arrays, reverse phase protein arrays (RPPA), microRNA, and somatic mutational signatures as -omics readouts.
+Across a collection of genes recurrently mutated in cancer, RNA sequencing tends to be the most effective predictor of mutation state.
+We find that one or more other data types for many of the genes are approximately equally effective predictors.
+Performance is more variable between mutations than that between data types for the same mutation, and there is little difference between the top data types.
+We also find that combining data types into a single multi-omics model provides little or no improvement in predictive ability over the best individual data type.
+
+#### Conclusions
+
+Based on our results, for the design of studies focused on the functional outcomes of cancer mutations, there are often multiple -omics types that can serve as effective readouts, although gene expression seems to be a reasonable default option.

content/32.introduction.md

@@ -0,0 +1,33 @@
+### Background
+
+Although cancer can be initiated and driven by many different genetic alterations, these tend to converge on a limited number of pathways or signaling processes [@doi:10.1016/j.cell.2018.03.035].
+As driver mutation status alone confers limited prognostic information, a comprehensive understanding of how diverse genetic alterations perturb central pathways is vital to precision medicine and biomarker identification efforts [@doi:10.7554/eLife.39217; @doi:10.1038/ncomms12096].
+While many methods exist to distinguish driver mutations from passenger mutations based on genomic sequence characteristics [@doi:10.1073/pnas.1616440113; @doi:10.1038/s41467-019-11284-9; @doi:10.1371/journal.pcbi.1006658], until recently it has been a challenge to connect driver mutations to downstream changes in gene expression and cellular function within individual tumor samples.
+
+The Cancer Genome Atlas (TCGA) Pan-Cancer Atlas provides uniformly processed, multi-platform -omics measurements across tens of thousands of samples from 33 cancer types [@doi:10.1038/ng.2764].
+Enabled by this publicly available data, a growing body of work on linking the presence of driving genetic alterations in cancer to downstream gene expression changes has emerged.
+Recent studies have considered Ras pathway alteration status in colorectal cancer [@doi:10.1158/1078-0432.CCR-13-1943], alteration status across many cancer types in Ras genes [@doi:10.1016/j.celrep.2018.03.046; @doi:10.1093/bib/bbaa258], _TP53_ [@doi:10.1016/j.celrep.2018.03.076], and _PIK3CA_ [@doi:10.1371/journal.pone.0241514], and alteration status across cancer types in frequently mutated genes [@doi:10.1186/s13059-020-02021-3].
+More broadly, other groups have drawn on similar ideas to distinguish between the functional effects of different alterations in the same driver gene [@doi:10.1101/2020.06.02.128850], to link alterations with similar gene expression signatures within cancer types [@doi:10.1142/9789811215636_0031], and to identify trans-acting expression quantitative trait loci (trans-eQTLs) in germline genetic studies [@doi:10.1101/2020.05.07.083386].
+
+These studies share a common thread: they each combine genomic (point mutation and copy number variation) data with transcriptomic (RNA sequencing) data within samples to interrogate the functional effects of genetic variation.
+RNA sequencing is ubiquitous and cheap, and its experimental and computational methods are relatively mature, making it a vital tool for generating insight into cancer pathology [@doi:10.1038/nrg.2017.96].
+Some driver mutations, however, are known to act indirectly on gene expression through varying mechanisms.
+For example, oncogenic _IDH1_ and _IDH2_ mutations in glioma have been shown to interfere with histone demethylation, which results in increased DNA methylation and blocked cell differentiation [@doi:10.1016/j.ccr.2010.03.017; @doi:10.1093/jnci/djq497; @doi:10.1056/NEJMoa0808710; @doi:10.1038/nature10860].
+Other genes implicated in aberrant DNA methylation in cancer include the TET family of genes [@doi:10.1016/j.tig.2014.07.005] and _SETD2_ [@doi:10.1101/cshperspect.a026468].
+Certain driver mutations, such as those in DNA damage repair genes, may lead to detectable patterns of somatic mutation [@doi:10.1038/nrg3729].
+Additionally, correlation between gene expression and protein abundance in cancer cell lines is limited, and proteomics data could correspond more directly to certain cancer phenotypes and pathway perturbations [@doi:10.1016/j.cell.2019.12.023].
+In these contexts and others, integrating different data modalities or combining multiple data modalities could be more effective than relying solely on gene expression as a functional signature.
+
+Here, we compare -omics data types profiled in the TCGA Pan-Cancer Atlas to evaluate use as a multivariate functional readout of genetic alterations in cancer.
+We focus on gene expression (RNA sequencing data), DNA methylation (27K and 450K probe chips), reverse phase protein array (RPPA), microRNA expression, and mutational signatures data [@doi:10.1038/s41586-020-1943-3] as possible readouts.
+Prior studies have identified univariate correlations of CpG site methylation [@doi:10.1371/journal.pcbi.1005840; @doi:10.1186/s12920-018-0425-z] and correlations of RPPA protein profiles [@doi:10.1186/s13073-018-0591-9] with the presence or absence of certain driver mutations.
+Other relevant past work includes linking point mutations and copy number variants (CNVs) with changes in methylation and expression at individual genes [@doi:10.1093/bioinformatics/btr019; @doi:10.1093/bib/bbw037] and identifying functional modules that are perturbed by somatic mutations [@doi:10.1093/bioinformatics/btq182; @doi:10.1038/ncomms9554].
+However, direct comparison among different data types for this application is lacking, particularly in the multivariate case where we consider changes to -omics-derived gene signatures rather than individual genes in isolation.
+
+We select a collection of potential cancer drivers with varying functions and roles in cancer development.
+We use mutation status in these genes as labels to train classifiers, using each of the data types listed as training data, in a pan-cancer setting; we follow similar methods to the elastic net logistic regression approach described in Way et al. 2018 [@doi:10.1016/j.celrep.2018.03.046] and Way et al. 2020 [@doi:10.1186/s13059-020-02021-3].
+We show that there is considerable predictive signal for many genes relative to a cancer-type corrected baseline and that gene expression tends to provide good predictions of mutation state across most genes.
+Surprisingly, we find that for a variety of genes, multiple data types are approximately equally effective predictors.
+We observe similar results for pan-cancer survival prediction across the same data types with little separation between the top-performing data types.
+In addition, we observe that combining data types into a single multi-omics model for mutation prediction provides little, if any, performance benefit over the most performant model using a single data type.
+Our results will help to inform the design of future functional genomics studies in cancer, suggesting that for many strong drivers with clear functional signatures, different -omics measurements can provide similar information content.