Matthew Ung

Big Data Bioinformatics

Recent technological advances allow for high throughput profiling of biological systems in a cost‐efficient manner. The low cost of data generation is leading us to the “big data” era. The availability of big data provides unprecedented opportunities but also raises new challenges for data mining and analysis. In this review, we introduce key concepts in the analysis of big data, including both “machine learning” algorithms as well as “unsupervised” and “supervised” examples of each. We note packages for the R programming language that are available to perform machine learning analyses. In addition to programming based solutions, we review webservers that allow users with limited or no programming background to perform these analyses on large data compendia. J. Cell. Physiol. 229: 1896–1900, 2014.

UNSUPERVISED FEATURE CONSTRUCTION AND KNOWLEDGE EXTRACTION FROM GENOME-WIDE ASSAYS OF BREAST CANCER WITH DENOISING AUTOENCODERS

Big data bring new opportunities for methods that efficiently summarize and automatically extract knowledge from such compendia. While both supervised learning algorithms and unsupervised clustering algorithms have been successfully applied to biological data, they are either dependent on known biology or limited to discerning the most significant signals in the data. Here we present denoising autoencoders (DAs), which employ a data-defined learning objective independent of known biology, as a method to identify and extract complex patterns from genomic data. We evaluate the performance of DAs by applying them to a large collection of breast cancer gene expression data. Results show that DAs successfully construct features that contain both clinical and molecular information. There are features that represent tumor or normal samples, estrogen receptor (ER) status, and molecular subtypes. Features constructed by the autoencoder generalize to an independent dataset collected using a distinct experimental platform. By integrating data from ENCODE for feature interpretation, we discover a feature representing ER status through association with key transcription factors in breast cancer. We also identify a feature highly predictive of patient survival and it is enriched by FOXM1 signaling pathway. The features constructed by DAs are often bimodally distributed with one peak near zero and another near one, which facilitates discretization. In summary, we demonstrate that DAs effectively extract key biological principles from gene expression data and summarize them into constructed features with convenient properties.

E2F4 regulatory program predicts patient survival prognosis in breast cancer

IntroductionGenetic and molecular signatures have been incorporated into cancer prognosis prediction and treatment decisions with good success over the past decade. Clinically, these signatures are usually used in early-stage cancers to evaluate whether they require adjuvant therapy following surgical resection. A molecular signature that is prognostic across more clinical contexts would be a useful addition to current signatures.MethodsWe defined a signature for the ubiquitous tissue factor, E2F4, based on its shared target genes in multiple tissues. These target genes were identified by chromatin immunoprecipitation sequencing (ChIP-seq) experiments using a probabilistic method. We then computationally calculated the regulatory activity score (RAS) of E2F4 in cancer tissues, and examined how E2F4 RAS correlates with patient survival.ResultsGenes in our E2F4 signature were 21-fold more likely to be correlated with breast cancer patient survival time compared to randomly selected genes. Using eight independent breast cancer datasets containing over 1,900 unique samples, we stratified patients into low and high E2F4 RAS groups. E2F4 activity stratification was highly predictive of patient outcome, and our results remained robust even when controlling for many factors including patient age, tumor size, grade, estrogen receptor (ER) status, lymph node (LN) status, whether the patient received adjuvant therapy, and the patient’s other prognostic indices such as Adjuvant! and the Nottingham Prognostic Index scores. Furthermore, the fractions of samples with positive E2F4 RAS vary in different intrinsic breast cancer subtypes, consistent with the different survival profiles of these subtypes.ConclusionsWe defined a prognostic signature, the E2F4 regulatory activity score, and showed it to be significantly predictive of patient outcome in breast cancer regardless of treatment status and the states of many other clinicopathological variables. It can be used in conjunction with other breast cancer classification methods such as Oncotype DX to improve clinical outcome prediction.

Effect of estrogen receptor α binding on functional DNA methylation in breast cancer

Epigenetic modifications introduce an additional layer of regulation that drastically expands the instructional capability of the human genome. The regulatory consequences of DNA methylation is context dependent; it can induce, enhance, and suppress gene expression, or have no effect on gene regulation. Therefore, it is essential to account for the genomic location of its occurrence and the protein factors it associates with to improve our understanding of its function and effects. Here, we use ENCODE ChIP-seq and DNase I hypersensitivity data, along with large-scale breast cancer genomic data from The Cancer Genome Atlas (TCGA) to computationally dissect the intricacies of DNA methylation in regulation of cancer transcriptomes. In particular, we identified a relationship between estrogen receptor α (ERα) activity and DNA methylation patterning in breast cancer. We found compelling evidence that methylation status of DNA sequences at ERα binding sites is tightly coupled with ERα activity. Furthermore, we predicted several transcription factors including FOXA1, GATA1, and SUZ12 to be associated with breast cancer by examining the methylation status of their binding sites in breast cancer. Lastly, we determine that methylated CpGs highly correlated with gene expression are enriched in regions 1kb or more downstream of TSSs, suggesting more significant regulatory roles for CpGs distal to gene TSSs. Our study provides novel insights into the role of ERα in breast cancers.

An approach for determining and measuring network hierarchy applied to comparing the phosphorylome and the regulome.

Many biological networks naturally form a hierarchy with a preponderance of downward information flow. In this study, we define a score to quantify the degree of hierarchy in a network and develop a simulated-annealing algorithm to maximize the hierarchical score globally over a network. We apply our algorithm to determine the hierarchical structure of the phosphorylome in detail and investigate the correlation between its hierarchy and kinase properties. We also compare it to the regulatory network, finding that the phosphorylome is more hierarchical than the regulome.

Integrative analysis of survival-associated gene sets in breast cancer

BackgroundPatient gene expression information has recently become a clinical feature used to evaluate breast cancer prognosis. The emergence of prognostic gene sets that take advantage of these data has led to a rich library of information that can be used to characterize the molecular nature of a patient’s cancer. Identifying robust gene sets that are consistently predictive of a patient’s clinical outcome has become one of the main challenges in the field.MethodsWe inputted our previously established BASE algorithm with patient gene expression data and gene sets from MSigDB to develop the gene set activity score (GSAS), a metric that quantitatively assesses a gene set’s activity level in a given patient. We utilized this metric, along with patient time-to-event data, to perform survival analyses to identify the gene sets that were significantly correlated with patient survival. We then performed cross-dataset analyses to identify robust prognostic gene sets and to classify patients by metastasis status. Additionally, we created a gene set network based on component gene overlap to explore the relationship between gene sets derived from MSigDB. We developed a novel gene set based on this network’s topology and applied the GSAS metric to characterize its role in patient survival.ResultsUsing the GSAS metric, we identified 120 gene sets that were significantly associated with patient survival in all datasets tested. The gene overlap network analysis yielded a novel gene set enriched in genes shared by the robustly predictive gene sets. This gene set was highly correlated to patient survival when used alone. Most interestingly, removal of the genes in this gene set from the gene pool on MSigDB resulted in a large reduction in the number of predictive gene sets, suggesting a prominent role for these genes in breast cancer progression.ConclusionsThe GSAS metric provided a useful medium by which we systematically investigated how gene sets from MSigDB relate to breast cancer patient survival. We used this metric to identify predictive gene sets and to construct a novel gene set containing genes heavily involved in cancer progression.

p53 and ΔNp63α Coregulate the Transcriptional and Cellular Response to TGFβ and BMP Signals

The TGFβ superfamily regulates a broad range of cellular processes, including proliferation, cell-fate specification, differentiation, and migration. Molecular mechanisms underlying this high degree of pleiotropy and cell-type specificity are not well understood. The TGFβ family is composed of two branches: (i) TGFβs, activins, and nodals, which signal through SMAD2/3, and (ii) bone morphogenetic proteins (BMP), which signal through SMAD1/5/8. SMADs have weak DNA-binding affinity and rely on coactivators and corepressors to specify their transcriptional outputs. This report reveals that p53 and ΔNp63α act as transcriptional partners for SMAD proteins and thereby influence cellular responses to TGFβ and BMPs. Suppression of p53 or overexpression of ΔNp63α synergistically enhance BMP-induced transcription. Mechanistically, p53 and ΔNp63α physically interact with SMAD1/5/8 proteins and co-occupy the promoter region of inhibitor of differentiation (ID2), a prosurvival BMP target gene. Demonstrating further convergence of these pathways, TGFβ-induced canonical BMP regulated transcription in a ΔNp63α- and p53-dependent manner. Furthermore, bioinformatic analyses revealed that SMAD2/3 and ΔNp63α coregulate a significant number of transcripts involved in the regulation of epithelial-to-mesenchymal transition. Thus, p53 and ΔNp63α are transcriptional partners for a subset of TGFβ- and BMP-regulated SMAD target genes in the mammary epithelium. Collectively, these results establish an integrated gene network of SMADs, p53, and ΔNp63α that contribute to EMT and metastasis. Implications: This study identifies aberrant BMP activation as a result of p53 mutation or ΔNp63α expression. Mol Cancer Res; 13(4); 732–42. ©2015 AACR.

Transcription Factor Binding Profiles Reveal Cyclic Expression of Human Protein-coding Genes and Non- coding RNAs

Cell cycle is a complex and highly supervised process that must proceed with regulatory precision to achieve successful cellular division. Despite the wide application, microarray time course experiments have several limitations in identifying cell cycle genes. We thus propose a computational model to predict human cell cycle genes based on transcription factor (TF) binding and regulatory motif information in their promoters. We utilize ENCODE ChIP-seq data and motif information as predictors to discriminate cell cycle against non-cell cycle genes. Our results show that both the trans- TF features and the cis- motif features are predictive of cell cycle genes, and a combination of the two types of features can further improve prediction accuracy. We apply our model to a complete list of GENCODE promoters to predict novel cell cycle driving promoters for both protein-coding genes and non-coding RNAs such as lincRNAs. We find that a similar percentage of lincRNAs are cell cycle regulated as protein-coding genes, suggesting the importance of non-coding RNAs in cell cycle division. The model we propose here provides not only a practical tool for identifying novel cell cycle genes with high accuracy, but also new insights on cell cycle regulation by TFs and cis-regulatory elements.

MYC Mediates mRNA Cap Methylation of Canonical Wnt/beta-catenin Signaling Transcripts by Recruiting CDK7 and RNA Methyltransferase

MYC is a pleiotropic transcription factor that activates and represses a wide range of target genes and is frequently deregulated in human tumors. While much is known about the role of MYC in transcriptional activation and repression, MYC can also regulate mRNA cap methylation through a mechanism that has remained poorly understood. Here, it is reported that MYC enhances mRNA cap methylation of transcripts globally, specifically increasing mRNA cap methylation of genes involved in Wnt/β-catenin signaling. Elevated mRNA cap methylation of Wnt signaling transcripts in response to MYC leads to augmented translational capacity, elevated protein levels, and enhanced Wnt signaling activity. Mechanistic evidence indicates that MYC promotes recruitment of RNA methyltransferase (RNMT) to Wnt signaling gene promoters by enhancing phosphorylation of serine 5 on the RNA polymerase II carboxy-terminal domain, mediated in part through an interaction between the TIP60 acetyltransferase complex and TFIIH. Implications: MYC enhances mRNA cap methylation above and beyond transcriptional induction. Mol Cancer Res; 15(2); 213–24. ©2016 AACR.

Regulators Associated with Clinical Outcomes Revealed by DNA Methylation Data in Breast Cancer

The regulatory architecture of breast cancer is extraordinarily complex and gene misregulation can occur at many levels, with transcriptional malfunction being a major cause. This dysfunctional process typically involves additional regulatory modulators including DNA methylation. Thus, the interplay between transcription factor (TF) binding and DNA methylation are two components of a cancer regulatory interactome presumed to display correlated signals. As proof of concept, we performed a systematic motif-based in silico analysis to infer all potential TFs that are involved in breast cancer prognosis through an association with DNA methylation changes. Using breast cancer DNA methylation and clinical data derived from The Cancer Genome Atlas (TCGA), we carried out a systematic inference of TFs whose misregulation underlie different clinical subtypes of breast cancer. Our analysis identified TFs known to be associated with clinical outcomes of p53 and ER (estrogen receptor) subtypes of breast cancer, while also predicting new TFs that may also be involved. Furthermore, our results suggest that misregulation in breast cancer can be caused by the binding of alternative factors to the binding sites of TFs whose activity has been ablated. Overall, this study provides a comprehensive analysis that links DNA methylation to TF binding to patient prognosis.