Kiley Graim

Inferring causal molecular networks: empirical assessment through a community-based effort

It remains unclear whether causal, rather than merely correlational, relationships in molecular networks can be inferred in complex biological settings. Here we describe the HPN-DREAM network inference challenge, which focused on learning causal influences in signaling networks. We used phosphoprotein data from cancer cell lines as well as in silico data from a nonlinear dynamical model. Using the phosphoprotein data, we scored more than 2,000 networks submitted by challenge participants. The networks spanned 32 biological contexts and were scored in terms of causal validity with respect to unseen interventional data. A number of approaches were effective, and incorporating known biology was generally advantageous. Additional sub-challenges considered time-course prediction and visualization. Our results suggest that learning causal relationships may be feasible in complex settings such as disease states. Furthermore, our scoring approach provides a practical way to empirically assess inferred molecular networks in a causal sense.

A basal stem cell signature identifies aggressive prostate cancer phenotypes

Significance Aggressive cancers often possess functional and molecular traits characteristic of normal stem cells. It is unclear if aggressive phenotypes of prostate cancer molecularly resemble normal stem cells residing within the human prostate. Here, we transcriptionally profiled epithelial populations from the human prostate and show that aggressive prostate cancer is enriched for a prostate basal stem cell signature. Within prostate cancer metastases, histological subtypes had varying enrichment of the stem cell signature, with small cell neuroendocrine carcinoma being the most stem cell-like. We further found that small cell neuroendocrine carcinoma and the prostate basal stem cell share a common transcriptional program. Targeting normal stem cell transcriptional programs may provide a new strategy for treating advanced prostate cancer. Evidence from numerous cancers suggests that increased aggressiveness is accompanied by up-regulation of signaling pathways and acquisition of properties common to stem cells. It is unclear if different subtypes of late-stage cancer vary in stemness properties and whether or not these subtypes are transcriptionally similar to normal tissue stem cells. We report a gene signature specific for human prostate basal cells that is differentially enriched in various phenotypes of late-stage metastatic prostate cancer. We FACS-purified and transcriptionally profiled basal and luminal epithelial populations from the benign and cancerous regions of primary human prostates. High-throughput RNA sequencing showed the basal population to be defined by genes associated with stem cell signaling programs and invasiveness. Application of a 91-gene basal signature to gene expression datasets from patients with organ-confined or hormone-refractory metastatic prostate cancer revealed that metastatic small cell neuroendocrine carcinoma was molecularly more stem-like than either metastatic adenocarcinoma or organ-confined adenocarcinoma. Bioinformatic analysis of the basal cell and two human small cell gene signatures identified a set of E2F target genes common between prostate small cell neuroendocrine carcinoma and primary prostate basal cells. Taken together, our data suggest that aggressive prostate cancer shares a conserved transcriptional program with normal adult prostate basal stem cells.

Combining heterogeneous data sources for accurate functional annotation of proteins

Combining heterogeneous sources of data is essential for accurate prediction of protein function. The task is complicated by the fact that while sequence-based features can be readily compared across species, most other data are species-specific. In this paper, we present a multi-view extension to GOstruct, a structured-output framework for function annotation of proteins. The extended framework can learn from disparate data sources, with each data source provided to the framework in the form of a kernel. Our empirical results demonstrate that the multi-view framework is able to utilize all available information, yielding better performance than sequence-based models trained across species and models trained from collections of data within a given species. This version of GOstruct participated in the recent Critical Assessment of Functional Annotations (CAFA) challenge; since then we have significantly improved the natural language processing component of the method, which now provides performance that is on par with that provided by sequence information. The GOstruct framework is available for download at http://strut.sourceforge.net.

Genomic evidence of widespread admixture from polar bears into brown bears during the last ice age

Abstract Recent genomic analyses have provided substantial evidence for past periods of gene flow from polar bears (Ursus maritimus) into Alaskan brown bears (Ursus arctos), with some analyses suggesting a link between climate change and genomic introgression. However, because it has mainly been possible to sample bears from the present day, the timing, frequency, and evolutionary significance of this admixture remains unknown. Here, we analyze genomic DNA from three additional and geographically distinct brown bear populations, including two that lived temporally close to the peak of the last ice age. We find evidence of admixture in all three populations, suggesting that admixture between these species has been common in their recent evolutionary history. In addition, analyses of ten fossil bears from the now‐extinct Irish population indicate that admixture peaked during the last ice age, whereas brown bear and polar bear ranges overlapped. Following this peak, the proportion of polar bear ancestry in Irish brown bears declined rapidly until their extinction. Our results support a model in which ice age climate change created geographically widespread conditions conducive to admixture between polar bears and brown bears, as is again occurring today. We postulate that this model will be informative for many admixing species pairs impacted by climate change. Our results highlight the power of paleogenomics to reveal patterns of evolutionary change that are otherwise masked in contemporary data.

Revealing cancer subtypes with higher-order correlations applied to imaging and omics data

BackgroundPatient stratification to identify subtypes with different disease manifestations, severity, and expected survival time is a critical task in cancer diagnosis and treatment. While stratification approaches using various biomarkers (including high-throughput gene expression measurements) for patient-to-patient comparisons have been successful in elucidating previously unseen subtypes, there remains an untapped potential of incorporating various genotypic and phenotypic data to discover novel or improved groupings.MethodsHere, we present HOCUS, a unified analytical framework for patient stratification that uses a community detection technique to extract subtypes out of sparse patient measurements. HOCUS constructs a patient-to-patient network from similarities in the data and iteratively groups and reconstructs the network into higher order clusters. We investigate the merits of using higher-order correlations to cluster samples of cancer patients in terms of their associations with survival outcomes.ResultsIn an initial test of the method, the approach identifies cancer subtypes in mutation data of glioblastoma, ovarian, breast, prostate, and bladder cancers. In several cases, HOCUS provides an improvement over using the molecular features directly to compare samples. Application of HOCUS to glioblastoma images reveals a size and location classification of tumors that improves over human expert-based stratification.ConclusionsSubtypes based on higher order features can reveal comparable or distinct groupings. The distinct solutions can provide biologically- and treatment-relevant solutions that are just as significant as solutions based on the original data.

Abstract PR02: Multiple Pathway Learning accurately predicts gene essentiality in the Cancer Cell Line Encyclopedia

We applied biologically-motivated feature transformations coupled with established machine learning methods to predict gene essentiality in CCLE cell line models. By leveraging additional large datasets, such as The Cancer Genome Atlas PanCancer12 data and MSigDB pathway definitions, we improved the robustness and biological interpretability of our models. We developed a multi-pathway learning (MPL) approach that associates a genetic pathway from MSigDB with a distinct kernel for use in a multiple kernel learning setting. We evaluated the performance of MPL compared to several other regression methods including random forests, kernel ridge regression, and elastic net linear models, We combined multiple approaches using an ensemble technique on the diverse set of predictors. We found that the best performing method was an ensemble combining MPL and random forest predictions. Both models utilized features derived from both gene expression and copy number data, the latter of which were filtered to those predicted as driver events in prior pan-cancer studies. The ensemble method was a joint winner in the recent DREAM 9 gene essentiality prediction challenge. MPL also demonstrated merit as a feature selector when used with other downstream methods. The ensemble performed best at predicting the essentiality of genes involved in cell cycle control (cyclins and cyclin-dependent kinases), protein degradation (proteasome complex), cell proliferation signaling (sonic hedgehog, Aurora-B, RAC1), apoptosis (RB1,TP53) and hypoxia response (VEGF, VHL). Many of the key genes in those pathways are known to be drivers of cancer progression, suggesting our method9s utility as a biomarker for detecting key tumorigenic events. The advantage of MPL is that mechanistically coherent gene sets are automatically selected as high scoring pathway kernels (HSPKs). We investigated whether the HSPKs identify cellular processes relevant to the loss of key genes. To do this, we inspected the HSPKs for a few of the most abundantly mutated genes in cancer. The MPL predictor for TP53 included the targets of this transcription factor as well as HSPKs involved in apoptosis, a cellular process regulated by TP53. The retinoblastoma gene (RB1) MPL predictor included RB1 targets as well as HSPKs involved in the regulation of histone deacetylase (HDAC) that interacts with RB1 to suppress DNA synthesis. These findings suggest trends in the MPL results could reveal a pathway-level view of the synthetic lethal architecture of cells. Such a map, that links patterns of pathway expression to potential genetic vulnerabilities, could provide an invaluable tool for exploring new avenues to target cancer cells. Citation Format: Vladislav Uzunangelov, Evan Paull, Sahil Chopra, Daniel Carlin, Adrian Bivol, Kyle Ellrott, Kiley Graim, Yulia Newton, Sam Ng, Artem Sokolov, Joshua Stuart. Multiple Pathway Learning accurately predicts gene essentiality in the Cancer Cell Line Encyclopedia. [abstract]. In: Proceedings of the AACR Special Conference on Computational and Systems Biology of Cancer; Feb 8-11 2015; San Francisco, CA. Philadelphia (PA): AACR; Cancer Res 2015;75(22 Suppl 2):Abstract nr PR02.

Abstract A2-64: A signature catalog to classify tumor mixtures: Application to recognition of metastatic disease in prostate cancer

Background: Molecular-based subtypes most certainly play a role in cancer progression and treatment. The recent results from the The Cancer Genome Atlas (TCGA) Pan-Cancer-12 analyses revealed connections between the cell of origin and patient outcomes. For example, bladder cancers were found to relate to three major Pan-Cancer integrative subtypes, with adeno-like and squamous-like bladder cancers associated with poorer prognosis than tumors with bladder-distinct profiles. Furthermore, both adeno-like lung and squamous-like bladder cancers were found to be associated with the most aggressive form of the disease. Methods: We are collecting a catalog of molecular signatures for each subtype found from Pan-Cancer analyses in TCGA and from relevant external datasets. Our goal is to map every tumor sample to one or more signatures in this collection using machine-learning methods. This mapping will allow us to predict the subclonal composition of primary tumor biopsies and to compare them to inferences from the variant allele frequency analysis, shedding light on the gene expression changes associated with key events in tumor evolution. As a pilot study, we compared signatures derived from metastatic prostate samples to subtypes of primary prostate tumors. Our goal is to test whether a molecular profile of metastatic disease can be recognized early on in primary tumors. To do so, we used unsupervised classification of mRNA expression profiles to define clusters of metastatic disease from external datasets as well as separately for primary tumors including the TCGA prostate adenocarcinoma dataset. We then performed an all-against-all comparison of signatures derived from metastatic subtypes to signatures derived from primary tumor subtypes. Results: The majority of metastatic tumors are most closely associated with one out of four primary subtypes, suggesting we have identified a possible primary signature associated with more aggressive disease. The finding is supported by enrichment analysis of clinical variables in the primary subtypes. Specifically, the primary subtype most often associated with the metastatic tumors have higher Gleason scores and higher tumor grade. In addition, several molecular pathways (e.g. BioCarta Vitamin D Receptor and KEGG Integrins in Angiogenesis pathways) and genes (e.g. MMP9, FGA, and LYZ) were found to be associated with the location of metastasis. Conclusions: Training molecular subtype recognizers may hold promise for detecting minor populations of subclones in primary and metastatic tumors. The subclone decomposition can be used to detect the presence of more aggressive disease that may resist standard treatment regimens. We are now expanding our signature catalog to include a more comprehensive collection and applying to additional subtypes of interest. We will make all datasets and signatures available through a mature version of the UCSC TumorMap portal. Citation Format: Kiley Graim, Yulia Newton, Adrian Bivol, Artem Sokolov, Kyle Ellrott, Robert Baertsch, Joshua Stuart. A signature catalog to classify tumor mixtures: Application to recognition of metastatic disease in prostate cancer. [abstract]. In: Proceedings of the AACR Special Conference on Translation of the Cancer Genome; Feb 7-9, 2015; San Francisco, CA. Philadelphia (PA): AACR; Cancer Res 2015;75(22 Suppl 1):Abstract nr A2-64.

Abstract 4177: Identification of pathways relevant for metastatic site prediction in prostate cancer

Proceedings: AACR Annual Meeting 2014; April 5-9, 2014; San Diego, CA Background and Significance: We address the problem of metastatic site prediction in prostate adenocarcinoma (PRAD), with a specific focus on identifying molecular pathways that are activated in association with the homing to a particular metastatic site. The approach can reveal the molecular mechanisms in metastatic cancer while also providing clues about potential drug targets. Further experimental validation of our findings may lead to the discovery of novel therapies for patients who are in the advanced stages of disease. Methods: We downloaded four PRAD datasets that contained met-site information from the Gene Expression Omnibus (GEO) and trained multi-class predictors on this set. The predictors were then evaluated on patient samples collected as part of the Stand Up To Cancer (SU2C) initiative. Standard normalization techniques were used to remove batch effects associated with non-biological factors such as the institution from which the materials were collected and/or assays conducted. We focused our attention on linear models due to their straightforward interpretation: higher weights indicate stronger association of the corresponding genomic features with a particular metastatic site. To identify pathways implicated by the relevant genomic features, we employed model regularization via group LASSO. This technique groups genes according to their pathway membership using the PathwayCommons database. The regularizer (penalty trading accurate classification with model complexity) sets the weights of an entire group to zero if those groups were uninformative for met-site prediction and non-zero otherwise. Results: We trained a multi-class linear predictor to recognize lymphatic node, liver and bone metastatic sites from gene expression data. The resulting model gave rise to two linear signatures: one that distinguished liver mets from the rest, and another that distinguished lymph node mets from the rest. The signatures were enriched for pathways commonly associated with liver development and liver progenitor cells, as well as pathways involved in integrin interactions on the cell surface. Based on the latter, we hypothesize that the up-regulation of particular integrin-signaling pathways may be responsible for driving the tendency of metastatic PRAD cells to prefer one site over another. We are currently in the process of investigating whether there is further evidence of this hypothesis in the SU2C data, as well as comparing group LASSO to other regularization techniques that also incorporate prior pathway information. Conclusion: We used linear methods to identify several pathways that may be responsible for localization of metastatic prostate adenocarcinoma cells to specific tissues. Our empirical results provide evidence that integrin-signaling may play a key role in this process. We are working on robustness evaluation of these findings, as well as experimental validation with our SU2C collaborators. Citation Format: Adrian Bivol, Kiley Graim, Evan Paull, Dan Carlin, Robert Baertsch, Artem Sokolov, Josh Stuart. Identification of pathways relevant for metastatic site prediction in prostate cancer. [abstract]. In: Proceedings of the 105th Annual Meeting of the American Association for Cancer Research; 2014 Apr 5-9; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2014;74(19 Suppl):Abstract nr 4177. doi:10.1158/1538-7445.AM2014-4177