Network Deconvolution Research Articles

Mixed Neural Network Approach for Temporal Sleep Stage Classification.

This paper proposes a practical approach to addressing limitations posed by using of single-channel electroencephalography (EEG) for sleep stage classification. EEG-based characterizations of sleep stage progression contribute the diagnosis and monitoring of the many pathologies of sleep. Several prior reports explored ways of automating the analysis of sleep EEG and of reducing the complexity of the data needed for reliable discrimination of sleep stages at lower cost in the home. However, these reports have involved recordings from electrodes placed on the cranial vertex or occiput, which are both uncomfortable and difficult to position. Previous studies of sleep stage scoring that used only frontal electrodes with a hierarchical decision tree motivated this paper, in which we have taken advantage of rectifier neural network for detecting hierarchical features and long short-term memory network for sequential data learning to optimize classification performance with single-channel recordings. After exploring alternative electrode placements, we found a comfortable configuration of a single-channel EEG on the forehead and have shown that it can be integrated with additional electrodes for simultaneous recording of the electro-oculogram. Evaluation of data from 62 people (with 494 hours sleep) demonstrated better performance of our analytical algorithm than is available from existing approaches with vertex or occipital electrode placements. Use of this recording configuration with neural network deconvolution promises to make clinically indicated home sleep studies practical.

DMirNet: Inferring direct microRNA-mRNA association networks

BackgroundMicroRNAs (miRNAs) play important regulatory roles in the wide range of biological processes by inducing target mRNA degradation or translational repression. Based on the correlation between expression profiles of a miRNA and its target mRNA, various computational methods have previously been proposed to identify miRNA-mRNA association networks by incorporating the matched miRNA and mRNA expression profiles. However, there remain three major issues to be resolved in the conventional computation approaches for inferring miRNA-mRNA association networks from expression profiles. 1) Inferred correlations from the observed expression profiles using conventional correlation-based methods include numerous erroneous links or over-estimated edge weight due to the transitive information flow among direct associations. 2) Due to the high-dimension-low-sample-size problem on the microarray dataset, it is difficult to obtain an accurate and reliable estimate of the empirical correlations between all pairs of expression profiles. 3) Because the previously proposed computational methods usually suffer from varying performance across different datasets, a more reliable model that guarantees optimal or suboptimal performance across different datasets is highly needed.ResultsIn this paper, we present DMirNet, a new framework for identifying direct miRNA-mRNA association networks. To tackle the aforementioned issues, DMirNet incorporates 1) three direct correlation estimation methods (namely Corpcor, SPACE, Network deconvolution) to infer direct miRNA-mRNA association networks, 2) the bootstrapping method to fully utilize insufficient training expression profiles, and 3) a rank-based Ensemble aggregation to build a reliable and robust model across different datasets.Our empirical experiments on three datasets demonstrate the combinatorial effects of necessary components in DMirNet. Additional performance comparison experiments show that DMirNet outperforms the state-of-the-art Ensemble-based model [1] which has shown the best performance across the same three datasets, with a factor of up to 1.29. Further, we identify 43 putative novel multi-cancer-related miRNA-mRNA association relationships from an inferred Top 1000 direct miRNA-mRNA association network.ConclusionsWe believe that DMirNet is a promising method to identify novel direct miRNA-mRNA relations and to elucidate the direct miRNA-mRNA association networks. Since DMirNet infers direct relationships from the observed data, DMirNet can contribute to reconstructing various direct regulatory pathways, including, but not limited to, the direct miRNA-mRNA association networks.Electronic supplementary materialThe online version of this article (doi:10.1186/s12918-016-0373-1) contains supplementary material, which is available to authorized users.

Inferring sparse networks for noisy transient processes.

Inferring causal structures of real world complex networks from measured time series signals remains an open issue. The current approaches are inadequate to discern between direct versus indirect influences (i.e., the presence or absence of a directed arc connecting two nodes) in the presence of noise, sparse interactions, as well as nonlinear and transient dynamics of real world processes. We report a sparse regression (referred to as the -min) approach with theoretical bounds on the constraints on the allowable perturbation to recover the network structure that guarantees sparsity and robustness to noise. We also introduce averaging and perturbation procedures to further enhance prediction scores (i.e., reduce inference errors), and the numerical stability of -min approach. Extensive investigations have been conducted with multiple benchmark simulated genetic regulatory network and Michaelis-Menten dynamics, as well as real world data sets from DREAM5 challenge. These investigations suggest that our approach can significantly improve, oftentimes by 5 orders of magnitude over the methods reported previously for inferring the structure of dynamic networks, such as Bayesian network, network deconvolution, silencing and modular response analysis methods based on optimizing for sparsity, transients, noise and high dimensionality issues.

Abstract 2381: Epithelial-stromal network deconvolution analysis reveals new targetable epithelial-stromal network dependencies in breast cancer

Abstract BACKGROUND: Epithelial-stromal crosstalk plays a critical role in breast carcinogenesis. However, the crosstalk networks involved in breast carcinogenesis are not yet known. Uncovering and understanding these networks may enable the development of diagnostics and therapeutics to target epithelial-stromal crosstalk networks for the prevention and treatment of breast cancer. Thus, we performed a genomewide epithelial-stromal co-expression and network deconvolution analysis to identify novel epithelial-stromal crosstalk partners in breast cancer. METHODS: We acquired five breast cancer laser capture microdissection data sets from the Gene Expression Omnibus. Paired epithelial-stromal tumor samples were obtained from 82 patients with invasive breast cancer (IBC) (164 samples) and paired epithelial-stromal normal samples were obtained from 41 patients (82 samples). IBC cases were divided into ER+ and ER- groups. To build epithelial-stromal co-expression networks, we computed the mutual information (MI) between all possible epithelial (E) and stromal (S) pairwise relationships (E-E, E-S, S-S). We then performed network deconvolution to prioritize direct dependencies. A subset of top-ranking dependencies was further evaluated by western blot in breast cancer cell lines and by immunohistochemistry on breast cancer tissue microarrays. RESULTS AND CONCLUSION: A massive number of total interactions were evaluated in our analyses (>400,000,000). The top interactions in ER+ IBC, ER- IBC and normal breast groups are known to functionally interact, validating our approach: E-E lipophilin B-mammaglobin interaction (ER+ IBC), S-S metallothionein 2A-metallothionein interaction (ER- IBC; MI = 0.95) and E-E RPS16-RPS3 interaction (normal breast; MI = 1). The functional gene set enrichment analysis using the DAVID bioinformatics tool revealed that the most enriched functional terms in ER+ and ER- IBC groups include: signal peptides, extracellular matrix proteins, and extracellular matrix:receptor interaction (all FDR < 0.00001). Using the ClusterONE algorithm, we identified a network cluster driven by Placental Growth Factor (PGF) in both ER+ and ER- IBC. PGF promotes angiogenesis, tumor growth, cell motility, and is expressed by 30-50% of primary breast cancers. We experimentally validated the positive association of PGF and GTF2H3 expression on tissue microarrays and in cell lines. PGF induces the GTF2H3 transcription factor expression in both MCF7 (ER+) and MDA231 (ER-) breast cancer cell lines. In conclusion, epithelial-stromal network deconvolution analysis represents a new approach for the prioritization of interacting epithelial and stromal crosstalk partners in breast cancer, providing new insights into network dependencies in breast carcinogenesis and potentially enabling the identification of new diagnostic and therapeutic targets. Citation Format: Octavian Bucur, Laleh Montaser-Kouhsari, Eun-Yeong Oh, Andrew H. Beck. Epithelial-stromal network deconvolution analysis reveals new targetable epithelial-stromal network dependencies in breast cancer. [abstract]. In: Proceedings of the 106th Annual Meeting of the American Association for Cancer Research; 2015 Apr 18-22; Philadelphia, PA. Philadelphia (PA): AACR; Cancer Res 2015;75(15 Suppl):Abstract nr 2381. doi:10.1158/1538-7445.AM2015-2381

Hopfield neural network deconvolution for weak lensing measurement

Weak gravitational lensing has the potential to place tight constraints on the equation of the state of dark energy. However, this will only be possible if shear measurement methods can reach the required level of accuracy. We present a new method to measure the ellipticity of galaxies used in weak lensing surveys. The method makes use of direct deconvolution of the data by the total Point Spread Function (PSF). We adopt a linear algebra formalism that represents the PSF as a Toeplitz matrix. This allows us to solve the convolution equation by applying the Hopfield Neural Network iterative scheme. The ellipticity of galaxies in the deconvolved images are then measured using second order moments of the autocorrelation function of the images. To our knowledge, it is the first time full image deconvolution is used to measure weak lensing shear. We apply our method to the simulated weak lensing data proposed in the GREAT10 challenge and obtain a quality factor of Q=87. This result is obtained after applying image denoising to the data, prior to the deconvolution. The additive and multiplicative biases on the shear power spectrum are then +0.000009 and +0.0357, respectively.

Improving accuracy of protein contact prediction using balanced network deconvolution.

Residue contact map is essential for protein three-dimensional structure determination. But most of the current contact prediction methods based on residue co-evolution suffer from high false-positives as introduced by indirect and transitive contacts (i.e., residues A-B and B-C are in contact, but A-C are not). Built on the work by Feizi et al. (Nat Biotechnol 2013; 31:726-733), which demonstrated a general network model to distinguish direct dependencies by network deconvolution, this study presents a new balanced network deconvolution (BND) algorithm to identify optimized dependency matrix without limit on the eigenvalue range in the applied network systems. The algorithm was used to filter contact predictions of five widely used co-evolution methods. On the test of proteins from three benchmark datasets of the 9th critical assessment of protein structure prediction (CASP9), CASP10, and PSICOV (precise structural contact prediction using sparse inverse covariance estimation) database experiments, the BND can improve the medium- and long-range contact predictions at the L/5 cutoff by 55.59% and 47.68%, respectively, without additional central processing unit cost. The improvement is statistically significant, with a P-value < 5.93 × 10(-3) in the Student's t-test. A further comparison with the ab initio structure predictions in CASPs showed that the usefulness of the current co-evolution-based contact prediction to the three-dimensional structure modeling relies on the number of homologous sequences existing in the sequence databases. BND can be used as a general contact refinement method, which is freely available at: http://www.csbio.sjtu.edu.cn/bioinf/BND/.

Highly sensitive inference of time-delayed gene regulation by network deconvolution.

BackgroundGene regulatory network (GRN) is a fundamental topic in systems biology. The dynamics of GRN can shed light on the cellular processes, which facilitates the understanding of the mechanisms of diseases when the processes are dysregulated. Accurate reconstruction of GRN could also provide guidelines for experimental biologists. Therefore, inferring gene regulatory networks from high-throughput gene expression data is a central problem in systems biology. However, due to the inherent complexity of gene regulation, noise in measuring the data and the short length of time-series data, it is very challenging to reconstruct accurate GRNs. On the other hand, a better understanding into gene regulation could help to improve the performance of GRN inference. Time delay is one of the most important characteristics of gene regulation. By incorporating the information of time delays, we can achieve more accurate inference of GRN.ResultsIn this paper, we propose a method to infer time-delayed gene regulation based on cross-correlation and network deconvolution (ND). First, we employ cross-correlation to obtain the probable time delays for the interactions between each target gene and its potential regulators. Then based on the inferred delays, the technique of ND is applied to identify direct interactions between the target gene and its regulators. Experiments on real-life gene expression datasets show that our method achieves overall better performance than existing methods for inferring time-delayed GRNs.ConclusionBy taking into account the time delays among gene interactions, our method is able to infer GRN more accurately. The effectiveness of our method has been shown by the experiments on three real-life gene expression datasets of yeast. Compared with other existing methods which were designed for learning time-delayed GRN, our method has significantly higher sensitivity without much reduction of specificity.

Characterization of microbial associations in human oral microbiome.

Microorganisms interact with each other within a community. Within the same community, some microorganisms tend to co-exist, whereas some others tend to avoid each other. The association among microorganisms can be revealed by computing the correlation between their abundance patterns that are measured through metagenomic sequencing across multiple communities. In this paper, we built an association network among microorganisms from the human oral microbiome. To improve its accuracy, we adopted a network deconvolution algorithm to filter out indirect associations, and we used an ensemble of three correlation measures to filter out the false-positive associations. When applying on the metagenomic data from human oral samples, experimental results showed that phylogenetically close microorganisms formed highly correlated network clusters. Additionally, most of the identified mutually exclusive associations were related to the order Lactobacillales.

Network deconvolution as a general method to distinguish direct dependencies in networks

Recognizing direct relationships between variables connected in a network is a pervasive problem in biological, social and information sciences as correlation-based networks contain numerous indirect relationships. Here we present a general method for inferring direct effects from an observed correlation matrix containing both direct and indirect effects. We formulate the problem as the inverse of network convolution, and introduce an algorithm that removes the combined effect of all indirect paths of arbitrary length in a closed-form solution by exploiting eigen-decomposition and infinite-series sums. We demonstrate the effectiveness of our approach in several network applications: distinguishing direct targets in gene expression regulatory networks; recognizing directly-interacting amino-acid residues for protein structure prediction from sequence alignments; and distinguishing strong collaborations in co-authorship social networks using connectivity information alone.

Synthetic microarray data generation with RANGE and NEMO

For testing and sensitivity analysis purposes, it is beneficial to have known transcription networks of sufficient size and variability during development of microarray data and network deconvolution algorithms. Description of such networks in a simple language translatable to Systems Biology Markup Language would allow generation of model data for the networks. Described herein is software (RANGE: RAndom Network GEnerator) to generate large random transcription networks in the NEMO (NEtwork MOtif) language. NEMO is recognized by a grammar for transcription network motifs using lex and yacc to output Systems Biology Markup Language models for either specified or randomized gene input functions. These models of known networks may be input to a biochemical simulator, allowing the generation of synthetic microarray data. http://range.sourceforge.net

ARACNE: An Algorithm for the Reconstruction of Gene Regulatory Networks in a Mammalian Cellular Context

BackgroundElucidating gene regulatory networks is crucial for understanding normal cell physiology and complex pathologic phenotypes. Existing computational methods for the genome-wide "reverse engineering" of such networks have been successful only for lower eukaryotes with simple genomes. Here we present ARACNE, a novel algorithm, using microarray expression profiles, specifically designed to scale up to the complexity of regulatory networks in mammalian cells, yet general enough to address a wider range of network deconvolution problems. This method uses an information theoretic approach to eliminate the majority of indirect interactions inferred by co-expression methods.ResultsWe prove that ARACNE reconstructs the network exactly (asymptotically) if the effect of loops in the network topology is negligible, and we show that the algorithm works well in practice, even in the presence of numerous loops and complex topologies. We assess ARACNE's ability to reconstruct transcriptional regulatory networks using both a realistic synthetic dataset and a microarray dataset from human B cells. On synthetic datasets ARACNE achieves very low error rates and outperforms established methods, such as Relevance Networks and Bayesian Networks. Application to the deconvolution of genetic networks in human B cells demonstrates ARACNE's ability to infer validated transcriptional targets of the cMYC proto-oncogene. We also study the effects of misestimation of mutual information on network reconstruction, and show that algorithms based on mutual information ranking are more resilient to estimation errors.ConclusionARACNE shows promise in identifying direct transcriptional interactions in mammalian cellular networks, a problem that has challenged existing reverse engineering algorithms. This approach should enhance our ability to use microarray data to elucidate functional mechanisms that underlie cellular processes and to identify molecular targets of pharmacological compounds in mammalian cellular networks.

Application of Euler Deconvolution and a Neural Network System as Interpretation Aids for Three Component Downhole Tem Data

The relationship between the secondary field due to a TEM source and a dipolar potential field has been documented (Dyck and West, 1984; Grant and West, 1965). In light of this relationship it is possible to utilise potential field interpretation aids when dealing with TEM data. Where only the axial component of the anomalous field is available, in the case of downhole TEM data, the implementation of these techniques is very restricted and may yield ambiguous model parameters, hence alternative techniques must be considered. Euler deconvolution may be used to obtain the approximate coordinates of the source even in the presence of an overburden response. A neural network system has been trained to output the dip and strike direction of a plate model using scaled field data as input. Utilising both the neural network and Euler deconvolution techniques, approximate model parameters may be computed directly from the field data. Synthetic examples are presented, showing that these techniques are applicable even in the presence of noise and conductive overburden. A field example from the Flying Doctor prospect is also presented. The computed model parameters were found to be consistent with previously published data (Cull, 1993).