A cell autonomous regulator of neuronal excitability modulates tau in Alzheimer's disease vulnerable neurons

Neurons from layer II of the entorhinal cortex (ECII) are the first to accumulate tau protein aggregates and degenerate during prodromal Alzheimer's disease (AD). Gaining insight into the molecular mechanisms underlying this vulnerability will help reveal genes and pathways at play during incipient stages of the disease. Here, we use a data-driven functional genomics approach to model ECII neurons in silico and identify the proto-oncogene DEK as a regulator of tau pathology. We show that epigenetic changes caused by Dek silencing alter activity-induced transcription, with major effects on neuronal excitability. This is accompanied by gradual accumulation of tau in the somatodendritic compartment of mouse ECII neurons in vivo, reactivity of surrounding microglia, and microglia-mediated neuron loss. These features are all characteristic of early AD. The existence of a cell-autonomous mechanism linking AD pathogenic mechanisms in the precise neuron type where the disease starts provides unique evidence that synaptic homeostasis dysregulation is of central importance in the onset of tau pathology in AD.

Interpretable neural architecture search and transfer learning for understanding CRISPR-Cas9 off-target enzymatic reactions

Finely tuned enzymatic pathways control cellular processes, and their dysregulation can lead to disease. Developing predictive and interpretable models for these pathways is challenging because of the complexity of the pathways and of the cellular and genomic contexts. Here we introduce Elektrum, a deep learning framework that addresses these challenges with data-driven and biophysically interpretable models for determining the kinetics of biochemical systems. First, it uses in vitro kinetic assays to rapidly hypothesize an ensemble of high-quality kinetically interpretable neural networks (KINNs) that predict reaction rates. It then employs a transfer learning step, where the KINNs are inserted as intermediary layers into deeper convolutional neural networks, fine-tuning the predictions for reaction-dependent in vivo outcomes. We apply Elektrum to predict CRISPR-Cas9 off-target editing probabilities and demonstrate that Elektrum achieves improved performance, regularizes neural network architectures and maintains physical interpretability.

Interpretable neural architecture search and transfer learning for understanding CRISPR/Cas9 off-target enzymatic reactions

Finely-tuned enzymatic pathways control cellular processes, and their dysregulation can lead to disease. Creating predictive and interpretable models for these pathways is challenging because of the complexity of the pathways and of the cellular and genomic contexts. Here we introduce Elektrum, a deep learning framework which addresses these challenges with data-driven and biophysically interpretable models for determining the kinetics of biochemical systems. First, it uses in vitro kinetic assays to rapidly hypothesize an ensemble of high-quality Kinetically Interpretable Neural Networks (KINNs) that predict reaction rates. It then employs a novel transfer learning step, where the KINNs are inserted as intermediary layers into deeper convolutional neural networks, fine-tuning the predictions for reaction-dependent in vivo outcomes. Elektrum makes effective use of the limited, but clean in vitro data and the complex, yet plentiful in vivo data that captures cellular context. We apply Elektrum to predict CRISPR-Cas9 off-target editing probabilities and demonstrate that Elektrum achieves state-of-the-art performance, regularizes neural network architectures, and maintains physical interpretability.

Oral mucosal breaks trigger anti-citrullinated bacterial and human protein antibody responses in rheumatoid arthritis

Periodontal disease is more common in individuals with rheumatoid arthritis (RA) who have detectable anti-citrullinated protein antibodies (ACPAs), implicating oral mucosal inflammation in RA pathogenesis. Here, we performed paired analysis of human and bacterial transcriptomics in longitudinal blood samples from RA patients. We found that patients with RA and periodontal disease experienced repeated oral bacteremias associated with transcriptional signatures of ISG15+HLADRhi and CD48highS100A2pos monocytes, recently identified in inflamed RA synovia and blood of those with RA flares. The oral bacteria observed transiently in blood were broadly citrullinated in the mouth, and their in situ citrullinated epitopes were targeted by extensively somatically hypermutated ACPAs encoded by RA blood plasmablasts. Together, these results suggest that (i) periodontal disease results in repeated breaches of the oral mucosa that release citrullinated oral bacteria into circulation, which (ii) activate inflammatory monocyte subsets that are observed in inflamed RA synovia and blood of RA patients with flares and (iii) activate ACPA B cells, thereby promoting affinity maturation and epitope spreading to citrullinated human antigens.

Molecular Characterization of Membranous Nephropathy

BACKGROUND

Molecular characterization of nephropathies may facilitate pathophysiologic insight, development of targeted therapeutics, and transcriptome-based disease classification. Although membranous nephropathy (MN) is a common cause of adult-onset nephrotic syndrome, the molecular pathways of kidney damage in MN require further definition.

METHODS

We applied a machine-learning framework to predict diagnosis on the basis of gene expression from the microdissected kidney tissue of participants in the Nephrotic Syndrome Study Network (NEPTUNE) cohort. We sought to identify differentially expressed genes between participants with MN versus those of other glomerulonephropathies across the NEPTUNE and European Renal cDNA Bank (ERCB) cohorts, to find MN-specific gene modules in a kidney-specific functional network, and to identify cell-type specificity of MN-specific genes using single-cell sequencing data from reference nephrectomy tissue.

RESULTS

Glomerular gene expression alone accurately separated participants with MN from those with other nephrotic syndrome etiologies. The top predictive classifier genes from NEPTUNE participants were also differentially expressed in the ERCB participants with MN. We identified a signature of 158 genes that are significantly differentially expressed in MN across both cohorts, finding 120 of these in a validation cohort. This signature is enriched in targets of transcription factor NF-κB. Clustering these MN-specific genes in a kidney-specific functional network uncovered modules with functional enrichments, including in ion transport, cell projection morphogenesis, regulation of adhesion, and wounding response. Expression data from reference nephrectomy tissue indicated 43% of these genes are most highly expressed by podocytes.

CONCLUSIONS

These results suggest that, relative to other glomerulonephropathies, MN has a distinctive molecular signature that includes upregulation of many podocyte-expressed genes, provides a molecular snapshot of MN, and facilitates insight into MN's underlying pathophysiology.

An activation to memory differentiation trajectory of tumor-infiltrating lymphocytes informs metastatic melanoma outcomes

There is a need for better classification and understanding of tumor-infiltrating lymphocytes (TILs). Here, we applied advanced functional genomics to interrogate 9,000 human tumors and multiple single-cell sequencing sets using benchmarked T cell states, comprehensive T cell differentiation trajectories, human and mouse vaccine responses, and other human TILs. Compared with other T cell states, enrichment of T memory/resident memory programs was observed across solid tumors. Trajectory analysis of single-cell melanoma CD8+ TILs also identified a high fraction of memory/resident memory-scoring TILs in anti-PD-1 responders, which expanded post therapy. In contrast, TILs scoring highly for early T cell activation, but not exhaustion, associated with non-response. Late/persistent, but not early activation signatures, prognosticate melanoma survival, and co-express with dendritic cell and IFN-γ response programs. These data identify an activation-like state associated to poor response and suggest successful memory conversion, above resuscitation of exhaustion, is an under-appreciated aspect of successful anti-tumoral immunity.

SynNotch-CAR T cells overcome challenges of specificity, heterogeneity, and persistence in treating glioblastoma

Treatment of solid cancers with chimeric antigen receptor (CAR) T cells is plagued by the lack of ideal target antigens that are both absolutely tumor specific and homogeneously expressed. We show that multi-antigen prime-and-kill recognition circuits provide flexibility and precision to overcome these challenges in the context of glioblastoma. A synNotch receptor that recognizes a specific priming antigen, such as the heterogeneous but tumor-specific glioblastoma neoantigen epidermal growth factor receptor splice variant III (EGFRvIII) or the central nervous system (CNS) tissue-specific antigen myelin oligodendrocyte glycoprotein (MOG), can be used to locally induce expression of a CAR. This enables thorough but controlled tumor cell killing by targeting antigens that are homogeneous but not absolutely tumor specific. Moreover, synNotch-regulated CAR expression averts tonic signaling and exhaustion, maintaining a higher fraction of the T cells in a naïve/stem cell memory state. In immunodeficient mice bearing intracerebral patient-derived xenografts (PDXs) with heterogeneous expression of EGFRvIII, a single intravenous infusion of EGFRvIII synNotch-CAR T cells demonstrated higher antitumor efficacy and T cell durability than conventional constitutively expressed CAR T cells, without off-tumor killing. T cells transduced with a synNotch-CAR circuit primed by the CNS-specific antigen MOG also exhibited precise and potent control of intracerebral PDX without evidence of priming outside of the brain. In summary, by using circuits that integrate recognition of multiple imperfect but complementary antigens, we improve the specificity, completeness, and persistence of T cells directed against glioblastoma, providing a general recognition strategy applicable to other solid tumors.

Abstract IA-12: Functional analysis of BRCA regulatory variants with deep learning models

Precision medicine in oncology depends on identifying variants that are functional and interpreting them in terms of cancer disease risk, treatment guidance, and prognosis. This is especially challenging for regulatory variants, which constitute the majority of the genome and control transcription and RNA processing. We use state-of-the-art deep learning models that predict the molecular impact of variants to systematically analyze genomes and SNPs and discover BRCA-associated SNPs that impact transcriptional and post-transcriptional regulation of gene expression, identify genes and biological processes implicated across tissues of impact, and reconcile these findings with prior genomic annotation-based analysis.

Citation Format: Olga Troyanskaya. Functional analysis of BRCA regulatory variants with deep learning models [abstract]. In: Proceedings of the AACR Virtual Special Conference on Artificial Intelligence, Diagnosis, and Imaging; 2021 Jan 13-14. Philadelphia (PA): AACR; Clin Cancer Res 2021;27(5_Suppl):Abstract nr IA-12.

Single cell transcriptomics identifies focal segmental glomerulosclerosis remission endothelial biomarker

To define cellular mechanisms underlying kidney function and failure, the KPMP analyzes biopsy tissue in a multicenter research network to build cell-level process maps of the kidney. This study aimed to establish a single cell RNA sequencing strategy to use cell-level transcriptional profiles from kidney biopsies in KPMP to define molecular subtypes in glomerular diseases. Using multiple sources of adult human kidney reference tissue samples, 22,268 single cell profiles passed KPMP quality control parameters. Unbiased clustering resulted in 31 distinct cell clusters that were linked to kidney and immune cell types using specific cell markers. Focusing on endothelial cell phenotypes, in silico and in situ hybridization methods assigned 3 discrete endothelial cell clusters to distinct renal vascular beds. Transcripts defining glomerular endothelial cells (GEC) were evaluated in biopsies from patients with 10 different glomerular diseases in the NEPTUNE and European Renal cDNA Bank (ERCB) cohort studies. Highest GEC scores were observed in patients with focal segmental glomerulosclerosis (FSGS). Molecular endothelial signatures suggested 2 distinct FSGS patient subgroups with α-2 macroglobulin (A2M) as a key downstream mediator of the endothelial cell phenotype. Finally, glomerular A2M transcript levels associated with lower proteinuria remission rates, linking endothelial function with long-term outcome in FSGS.

Abstract SY36-01: Decoding the biochemical, regulatory, and clinical impact of genomic mutations

A key challenge in precision medicine lies in decoding the complexity of the functional effects of genome variation to determine which mutations are functional and what are their biochemical and phenotypic consequences. This question is especially difficult for the 98% of the genome that is outside of exomes. To address this challenge we developed deep learning-based methods, DeepSEA and SeqWeaver, that predict the transcriptional and post-transcriptional effects of noncoding variants with single-nucleotide sensitivity. These methods not only classify mutations as likely to be associated with disease, they also provide specific biochemical consequences of each mutation, including on transcription factor and RBP binding, histone modification, and DNA accessibility. Building upon these findings, we created a deep learning-based framework, ExPecto, that can accurately predict, ab initio from DNA sequence alone, the tissue-specific transcriptional effects of mutations. These methods thus provide a platform for predicting and characterizing specific mutational impact for any genomic mutation, including those that are rare or that have not been observed. Applying these methods (available at hb.flatironinstitute.org) to whole genomes in a number of diseases, including cancer, autism, and heart disease, we demonstrate the contribution of de novo noncoding mutations and analyze the functional landscape of noncoding cancer-associated mutations.

Citation Format: Olga Troyanskaya. Decoding the biochemical, regulatory, and clinical impact of genomic mutations [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2019; 2019 Mar 29-Apr 3; Atlanta, GA. Philadelphia (PA): AACR; Cancer Res 2019;79(13 Suppl):Abstract nr SY36-01.

Bioinformatics Approaches to Profile the Tumor Microenvironment for Immunotherapeutic Discovery

In the microenvironment of a malignancy, tumor cells do not exist in isolation, but rather in a diverse ecosystem consisting not only of heterogeneous tumor-cell clones, but also normal cell types such as fibroblasts, vasculature, and an extensive pool of immune cells at numerous possible stages of activation and differentiation. This results in a complex interplay of diverse cellular signaling systems, where the immune cell component is now established to influence cancer progression and therapeutic response. It is experimentally difficult and laborious to comprehensively and systematically profile these distinct cell types from heterogeneous tumor samples in order to capitalize on potential therapeutic and biomarker discoveries. One emerging solution to address this challenge is to computationally extract cell-type specific information directly from bulk tumors. Such in silico approaches are advantageous because they can capture both the cell-type specific profiles and the tissue systems level of cell-cell interactions. Accurately and comprehensively predicting these patterns in tumors is an important challenge to overcome, not least given the success of immunotherapeutic drug treatment of several human cancers. This is especially challenging for subsets of closely related immune cell phenotypes with relatively small gene expression differences, which have critical functional distinctions. Here, we outline the existing and emerging novel bioinformatics strategies that can be used to profile the tumor immune landscape.

Single-cell analysis of progenitor cell dynamics and lineage specification in the human fetal kidney

The mammalian kidney develops through reciprocal interactions between the ureteric bud and the metanephric mesenchyme to give rise to the entire collecting system and the nephrons. Most of our knowledge of the developmental regulators driving this process arises from the study of gene expression and functional genetics in mice and other animal models. In order to shed light on human kidney development, we have used single-cell transcriptomics to characterize gene expression in different cell populations, and to study individual cell dynamics and lineage trajectories during development. Single-cell transcriptome analyses of 6414 cells from five individual specimens identified 11 initial clusters of specific renal cell types as defined by their gene expression profile. Further subclustering identifies progenitors, and mature and intermediate stages of differentiation for several renal lineages. Other lineages identified include mesangium, stroma, endothelial and immune cells. Novel markers for these cell types were revealed in the analysis, as were components of key signaling pathways driving renal development in animal models. Altogether, we provide a comprehensive and dynamic gene expression profile of the developing human kidney at the single-cell level.

Summary: New markers for specific cell types in the developing human kidney are identified and computational approaches infer developmental trajectories and interrogate the complex network of signaling pathways and cellular transitions.

Serum cytokine levels in breast cancer patients during neoadjuvant treatment with bevacizumab

A high concentration of circulating vascular endothelial growth factor (VEGF) in cancer patients is associated with an aggressive tumor phenotype. Here, serum levels of 27 cytokines and blood cell counts were assessed in breast cancer patients receiving neoadjuvant chemotherapy with or without bevacizumab (Bev) in a randomized cohort of 132 patients with non-metastatic HER2-negative tumors. Cytokine levels were determined prior to treatment and at various time-points. The cytotoxic chemotherapy regimen of fluorouracil, epirubicin, and cyclophosphamide (FEC) had a profound impact on both circulating white blood cells and circulating cytokine levels. At the end of FEC treatment, the global decrease in cytokine levels correlated with the drop in white blood cell counts and was significantly greater in the patients of the Bev arm for cytokines, such as VEGF-A, IL-12, IP-10 and IL-10. Among patients who received Bev, those with pathological complete response (pCR) exhibited significantly lower levels of VEGF-A, IFN-γ, TNF-α and IL-4 than patients without pCR. This effect was not observed in the chemotherapy-only arm. Certain circulating cytokine profiles were found to correlate with different immune cell types at the tumor site. For the Bev arm patients, the serum cytokine levels correlated with higher levels of cytotoxic T cells at the end of the therapy regimen, which was indicative of treatment response. The higher response rate for Bev-treated patients and stronger correlations between serum cytokine levels and infiltrating CD8T cells merits further investigation.

A loop-counting method for covariate-corrected low-rank biclustering of gene-expression and genome-wide association study data

A common goal in data-analysis is to sift through a large data-matrix and detect any significant submatrices (i.e., biclusters) that have a low numerical rank. We present a simple algorithm for tackling this biclustering problem. Our algorithm accumulates information about 2-by-2 submatrices (i.e., 'loops') within the data-matrix, and focuses on rows and columns of the data-matrix that participate in an abundance of low-rank loops. We demonstrate, through analysis and numerical-experiments, that this loop-counting method performs well in a variety of scenarios, outperforming simple spectral methods in many situations of interest. Another important feature of our method is that it can easily be modified to account for aspects of experimental design which commonly arise in practice. For example, our algorithm can be modified to correct for controls, categorical- and continuous-covariates, as well as sparsity within the data. We demonstrate these practical features with two examples; the first drawn from gene-expression analysis and the second drawn from a much larger genome-wide-association-study (GWAS).

IFNγ-Dependent Tissue-Immune Homeostasis Is Co-opted in the Tumor Microenvironment

Homeostatic programs balance immune protection and self-tolerance. Such mechanisms likely impact autoimmunity and tumor formation, respectively. How homeostasis is maintained and impacts tumor surveillance is unknown. Here, we find that different immune mononuclear phagocytes share a conserved steady-state program during differentiation and entry into healthy tissue. IFNγ is necessary and sufficient to induce this program, revealing a key instructive role. Remarkably, homeostatic and IFNγ-dependent programs enrich across primary human tumors, including melanoma, and stratify survival. Single-cell RNA sequencing (RNA-seq) reveals enrichment of homeostatic modules in monocytes and DCs from human metastatic melanoma. Suppressor-of-cytokine-2 (SOCS2) protein, a conserved program transcript, is expressed by mononuclear phagocytes infiltrating primary melanoma and is induced by IFNγ. SOCS2 limits adaptive anti-tumoral immunity and DC-based priming of T cells in vivo, indicating a critical regulatory role. These findings link immune homeostasis to key determinants of anti-tumoral immunity and escape, revealing co-opting of tissue-specific immune development in the tumor microenvironment.

Lymphocyte Invasion in IC10/Basal-Like Breast Tumors Is Associated with Wild-Type TP53

Lymphocytic infiltration is associated with better prognosis in several epithelial malignancies including breast cancer. The tumor suppressor TP53 is mutated in approximately 30% of breast adenocarcinomas, with varying frequency across molecular subtypes. In this study of 1,420 breast tumors, we tested for interaction between TP53 mutation status and tumor subtype determined by PAM50 and integrative cluster analysis. In integrative cluster 10 (IC10)/basal-like breast cancer, we identify an association between lymphocytic infiltration, determined by an expression score, and retention of wild-type TP53. The expression-derived score agreed with the degree of lymphocytic infiltration assessed by pathologic review, and application of the Nanodissect algorithm was suggestive of this infiltration being primarily of cytotoxic T lymphocytes (CTL). Elevated expression of this CTL signature was associated with longer survival in IC10/Basal-like tumors. These findings identify a new link between the TP53 pathway and the adaptive immune response in estrogen receptor (ER)-negative breast tumors, suggesting a connection between TP53 inactivation and failure of tumor immunosurveillance.

IMPLICATIONS

The association of lymphocytic invasion of ER-negative breast tumors with the retention of wild-type TP53 implies a novel protective connection between TP53 function and tumor immunosurveillance.

Abstract PR06: Lymphocyte invasion of basal breast tumors is associated with wild-type p53

The tumor suppressor p53 is mutated in approximately 30% of breast adenocarcinomas, but the clinical implications of p53 mutation depend on the subtype of breast tumor. Although p53 mutation is associated with worse outcome for patients with Luminal B and Her2 subtypes, the same is not true in Basal tumors, despite the fact that Basal tumors have the highest frequency of p53 mutation of any subtype. Lymphocytic infiltration is associated with better prognosis in several epithelial malignancies including triple-negative breast tumors. In this study we identify an association in Basal breast cancer between lymphocytic infiltration and the number of copies of wild-type p53 by combining gene expression, p53 sequencing, DNA copy number analysis, and semi-quantitative immune cell histology in 1,281 tumors. We use the Nanodissect algorithm to define signatures for T cell types and show that the infiltration consists primarily of cytotoxic T lymphocytes. Elevated expression of this gene signature is associated with better outcome in Basal tumors, and differential pathway activity analysis identifies significant activation of caveolin 1 signaling to TGF-beta in p53 wild-type Basal tumors. These findings identify a new link between the p53 pathway and beneficial adaptive immune response in breast tumors not responsive to the estrogen receptor, supporting a connection between a lack of p53 function and the failure of tumor immunosurveillance.

This abstract is also presented as Poster A35.

Citation Format: David A. Quigley, Charlie Vaske, Laxmi Silwa-Pandit, Ruth Dannenfelser, Anita Langerød, Hans Kristian M. Vollan, Olga Troyanskaya, Suet-Feung Chin, Carlos Caldas, Allan Balmain, Anne-Lise Børresen-Dale, Vessela Kristensen. Lymphocyte invasion of basal breast tumors is associated with wild-type p53. [abstract]. In: Abstracts: AACR Special Conference on Cellular Heterogeneity in the Tumor Microenvironment; 2014 Feb 26-Mar 1; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2015;75(1 Suppl):Abstract nr PR06. doi:10.1158/1538-7445.CHTME14-PR06

Investigating immunological pathways and diseases with a comprehensive compendium of human data (HUM8P.347)

The exponential growth of high throughput immunological data motivates the need to leverage this global effort to guide new investigation and to contextualize domain-specific results. Improved specificity and accuracy over currently utilized approaches such as ontological analysis would be advantageous. To develop such a resource (ImmuNet), we used immunological pathway-guided Bayesian integration of a comprehensive, heterogeneous compendium of human data to assemble functional relationship networks. The utility of ImmuNet was evaluated in various applications. We studied the cellular gene response signature to virus infection, using ImmuNet to guide confirmatory experiments into functionally relevant responses to different viruses. We also demonstrated that ImmuNet can accurately predict disease associated genes in the vicinity of GWAS loci and to explore processes and pathways underlying immune-mediated diseases. By unlocking immunological information captured within the global biomedical research effort, ImmuNet should be widely beneficial to future studies investigating the mechanisms of the human immune system and immunological disease. An interactive user-friendly web interface for investigators is publicly available at immunet.princeton.edu.

Systematic planning of genome-scale experiments in poorly studied species

Genome-scale datasets have been used extensively in model organisms to screen for specific candidates or to predict functions for uncharacterized genes. However, despite the availability of extensive knowledge in model organisms, the planning of genome-scale experiments in poorly studied species is still based on the intuition of experts or heuristic trials. We propose that computational and systematic approaches can be applied to drive the experiment planning process in poorly studied species based on available data and knowledge in closely related model organisms. In this paper, we suggest a computational strategy for recommending genome-scale experiments based on their capability to interrogate diverse biological processes to enable protein function assignment. To this end, we use the data-rich functional genomics compendium of the model organism to quantify the accuracy of each dataset in predicting each specific biological process and the overlap in such coverage between different datasets. Our approach uses an optimized combination of these quantifications to recommend an ordered list of experiments for accurately annotating most proteins in the poorly studied related organisms to most biological processes, as well as a set of experiments that target each specific biological process. The effectiveness of this experiment- planning system is demonstrated for two related yeast species: the model organism Saccharomyces cerevisiae and the comparatively poorly studied Saccharomyces bayanus. Our system recommended a set of S. bayanus experiments based on an S. cerevisiae microarray data compendium. In silico evaluations estimate that less than 10% of the experiments could achieve similar functional coverage to the whole microarray compendium. This estimation was confirmed by performing the recommended experiments in S. bayanus, therefore significantly reducing the labor devoted to characterize the poorly studied genome. This experiment-planning framework could readily be adapted to the design of other types of large-scale experiments as well as other groups of organisms.

Microarray expression experiments allow fast functional profiling of an organism's entire genome and significant efforts are devoted to analyzing the resulting data. Available genome sequences are also increasing quickly. However, it is unexplored how to use available functional genomics data to direct large-scale experiments in newly sequenced but poorly studied species. In this paper, we propose a strategy to systematically plan experimental treatments in the poorly studied species based on their model organism relatives. We consider both the accuracy of the datasets in capturing different biological processes and the redundancy between datasets. Quantifying the above information allows us to recommend a list of experimental treatments. We demonstrate the efficacy of this approach by designing, performing and evaluating S. bayanus microarray experiments using an available S. cerevisiae data repository. We show that this systematic planning process could reduce the labor in doing microarray experiments by 10 fold and achieve similar functional coverage.

Gene expression patterns in ovarian carcinomas

We used DNA microarrays to characterize the global gene expression patterns in surface epithelial cancers of the ovary. We identified groups of genes that distinguished the clear cell subtype from other ovarian carcinomas, grade I and II from grade III serous papillary carcinomas, and ovarian from breast carcinomas. Six clear cell carcinomas were distinguished from 36 other ovarian carcinomas (predominantly serous papillary) based on their gene expression patterns. The differences may yield insights into the worse prognosis and therapeutic resistance associated with clear cell carcinomas. A comparison of the gene expression patterns in the ovarian cancers to published data of gene expression in breast cancers revealed a large number of differentially expressed genes. We identified a group of 62 genes that correctly classified all 125 breast and ovarian cancer specimens. Among the best discriminators more highly expressed in the ovarian carcinomas were PAX8 (paired box gene 8), mesothelin, and ephrin-B1 (EFNB1). Although estrogen receptor was expressed in both the ovarian and breast cancers, genes that are coregulated with the estrogen receptor in breast cancers, including GATA-3, LIV-1, and X-box binding protein 1, did not show a similar pattern of coexpression in the ovarian cancers.

Missing value estimation methods for DNA microarrays

MOTIVATION

Gene expression microarray experiments can generate data sets with multiple missing expression values. Unfortunately, many algorithms for gene expression analysis require a complete matrix of gene array values as input. For example, methods such as hierarchical clustering and K-means clustering are not robust to missing data, and may lose effectiveness even with a few missing values. Methods for imputing missing data are needed, therefore, to minimize the effect of incomplete data sets on analyses, and to increase the range of data sets to which these algorithms can be applied. In this report, we investigate automated methods for estimating missing data.

RESULTS

We present a comparative study of several methods for the estimation of missing values in gene microarray data. We implemented and evaluated three methods: a Singular Value Decomposition (SVD) based method (SVDimpute), weighted K-nearest neighbors (KNNimpute), and row average. We evaluated the methods using a variety of parameter settings and over different real data sets, and assessed the robustness of the imputation methods to the amount of missing data over the range of 1--20% missing values. We show that KNNimpute appears to provide a more robust and sensitive method for missing value estimation than SVDimpute, and both SVDimpute and KNNimpute surpass the commonly used row average method (as well as filling missing values with zeros). We report results of the comparative experiments and provide recommendations and tools for accurate estimation of missing microarray data under a variety of conditions.

Missing value estimation methods for DNA microarrays

MOTIVATION

Gene expression microarray experiments can generate data sets with multiple missing expression values. Unfortunately, many algorithms for gene expression analysis require a complete matrix of gene array values as input. For example, methods such as hierarchical clustering and K-means clustering are not robust to missing data, and may lose effectiveness even with a few missing values. Methods for imputing missing data are needed, therefore, to minimize the effect of incomplete data sets on analyses, and to increase the range of data sets to which these algorithms can be applied. In this report, we investigate automated methods for estimating missing data.

RESULTS

We present a comparative study of several methods for the estimation of missing values in gene microarray data. We implemented and evaluated three methods: a Singular Value Decomposition (SVD) based method (SVDimpute), weighted K-nearest neighbors (KNNimpute), and row average. We evaluated the methods using a variety of parameter settings and over different real data sets, and assessed the robustness of the imputation methods to the amount of missing data over the range of 1--20% missing values. We show that KNNimpute appears to provide a more robust and sensitive method for missing value estimation than SVDimpute, and both SVDimpute and KNNimpute surpass the commonly used row average method (as well as filling missing values with zeros). We report results of the comparative experiments and provide recommendations and tools for accurate estimation of missing microarray data under a variety of conditions.