Ryo Yoshida

Statistical inference of transcriptional module-based gene networks from time course gene expression profiles by using state space models

MOTIVATION Statistical inference of gene networks by using time-course microarray gene expression profiles is an essential step towards understanding the temporal structure of gene regulatory mechanisms. Unfortunately, most of the current studies have been limited to analysing a small number of genes because the length of time-course gene expression profiles is fairly short. One promising approach to overcome such a limitation is to infer gene networks by exploring the potential transcriptional modules which are sets of genes sharing a common function or involved in the same pathway. RESULTS In this article, we present a novel approach based on the state space model to identify the transcriptional modules and module-based gene networks simultaneously. The state space model has the potential to infer large-scale gene networks, e.g. of order 10(3), from time-course gene expression profiles. Particularly, we succeeded in the identification of a cell cycle system by using the gene expression profiles of Saccharomyces cerevisiae in which the length of the time-course and number of genes were 24 and 4382, respectively. However, when analysing shorter time-course data, e.g. of length 10 or less, the parameter estimations of the state space model often fail due to overfitting. To extend the applicability of the state space model, we provide an approach to use the technical replicates of gene expression profiles, which are often measured in duplicate or triplicate. The use of technical replicates is important for achieving highly-efficient inferences of gene networks with short time-course data. The potential of the proposed method has been demonstrated through the time-course analysis of the gene expression profiles of human umbilical vein endothelial cells (HUVECs) undergoing growth factor deprivation-induced apoptosis. AVAILABILITY Supplementary Information and the software (TRANS-MNET) are available at http://daweb.ism.ac.jp/~yoshidar/software/ssm/.

Finding module-based gene networks with state-space models - Mining high-dimensional and short time-course gene expression data

This study explores some problems to analyze time-course gene expression data by state-space models (SSMs). One problem is regarding the methods of parameter estimation and determination of the dimension of the internal state variable. Although several methods have been applied, there are few literature studies which with to compare them. Thus, this paper gives a brief review of the existing literature that use the SSM to analyze the gene expression time-course data. Another problem is the identifiability of the model. If the parameters of SSMs are simply estimated without any constraints for parameter space, they lack identifiability. To identify a system uniquely, it requires a specific algorithm to estimate the parameters with some constraints. For that purpose, an identifiable form of SSMs and an algorithm for estimating parameters are derived. The last problem is the extraction of biological information by interpreting the estimated parameters, such as mechanism of gene regulations at the module level. For that one, this paper explores methods to extract further information using the estimated parameters, that is, reconstruction of a module network from time-course gene expression data

Epidermal Growth Factor Receptor Tyrosine Kinase Defines Critical Prognostic Genes of Stage I Lung Adenocarcinoma

Purpose To identify stage I lung adenocarcinoma patients with a poor prognosis who will benefit from adjuvant therapy. Patients and Methods Whole gene expression profiles were obtained at 19 time points over a 48-hour time course from human primary lung epithelial cells that were stimulated with epidermal growth factor (EGF) in the presence or absence of a clinically used EGF receptor tyrosine kinase (RTK)-specific inhibitor, gefitinib. The data were subjected to a mathematical simulation using the State Space Model (SSM). “Gefitinib-sensitive” genes, the expressional dynamics of which were altered by addition of gefitinib, were identified. A risk scoring model was constructed to classify high- or low-risk patients based on expression signatures of 139 gefitinib-sensitive genes in lung cancer using a training data set of 253 lung adenocarcinomas of North American cohort. The predictive ability of the risk scoring model was examined in independent cohorts of surgical specimens of lung cancer. Results The risk scoring model enabled the identification of high-risk stage IA and IB cases in another North American cohort for overall survival (OS) with a hazard ratio (HR) of 7.16 (P = 0.029) and 3.26 (P = 0.0072), respectively. It also enabled the identification of high-risk stage I cases without bronchioalveolar carcinoma (BAC) histology in a Japanese cohort for OS and recurrence-free survival (RFS) with HRs of 8.79 (P = 0.001) and 3.72 (P = 0.0049), respectively. Conclusion The set of 139 gefitinib-sensitive genes includes many genes known to be involved in biological aspects of cancer phenotypes, but not known to be involved in EGF signaling. The present result strongly re-emphasizes that EGF signaling status in cancer cells underlies an aggressive phenotype of cancer cells, which is useful for the selection of early-stage lung adenocarcinoma patients with a poor prognosis. Trial Registration The Gene Expression Omnibus (GEO) GSE31210

Estimating time-dependent gene networks from time series microarray data by dynamic linear models with Markov switching

In gene network estimation from time series microarray data, dynamic models such as differential equations and dynamic Bayesian networks assume that the network structure is stable through all time points, while the real network might changes its structure depending on time, affection of some shocks and so on. If the true network structure underlying the data changes at certain points, the fitting of the usual dynamic linear models fails to estimate the structure of gene network and we cannot obtain efficient information from data. To solve this problem, we propose a dynamic linear model with Markov switching for estimating time-dependent gene network structure from time series gene expression data. Using our proposed method, the network structure between genes and its change points are automatically estimated. We demonstrate the effectiveness of the proposed method through the analysis of Saccharomyces cerevisiae cell cycle time series data.

A mixed factors model for dimension reduction and extraction of a group structure in gene expression data

When we cluster tissue samples on the basis of genes, the number of observations to be grouped is much smaller than the dimension of feature vector. In such a case, the applicability of conventional model-based clustering is limited since the high dimensionality of feature vector leads to overfilling during the density estimation process. To overcome such difficulty, we attempt a methodological extension of the factor analysis. Our approach enables us not only to prevent from the occurrence of overfilling, but also to handle the issues of clustering, data compression and extracting a set of genes to be relevant to explain the group structure. The potential usefulness are demonstrated with the application to the leukemia dataset.

A STATE SPACE REPRESENTATION OF VAR MODELS WITH SPARSE LEARNING FOR DYNAMIC GENE NETWORKS

We propose a state space representation of vector autoregressive model and its sparse learning based on L1 regularization to achieve efficient estimation of dynamic gene networks based on time course microarray data. The proposed method can overcome drawbacks of the vector autoregressive model and state space model; the assumption of equal time interval and lack of separation ability of observation and systems noises in the former method and the assumption of modularity of network structure in the latter method. However, in a simple implementation the proposed model requires the calculation of large inverse matrices in a large number of times during parameter estimation process based on EM algorithm. This limits the applicability of the proposed method to a relatively small gene set. We thus introduce a new calculation technique for EM algorithm that does not require the calculation of inverse matrices. The proposed method is applied to time course microarray data of lung cells treated by stimulating EGF receptors and dosing an anticancer drug, Gefitinib. By comparing the estimated network with the control network estimated using non-treated lung cells, perturbed genes by the anticancer drug could be found, whose up- and down-stream genes in the estimated networks may be related to side effects of the anticancer drug.

PARAMETER ESTIMATION OF IN SILICO BIOLOGICAL PATHWAYS WITH PARTICLE FILTERING TOWARDS A PETASCALE COMPUTING

The aim of this paper is to demonstrate the potential power of large-scale particle filtering for the parameter estimations of in silico biological pathways where time course measurements of biochemical reactions are observable. The method of particle filtering has been a popular technique in the field of statistical science, which approximates posterior distributions of model parameters of dynamic system by using sequentially-generated Monte Carlo samples. In order to apply the particle filtering to system identifications of biological pathways, it is often needed to explore the posterior distributions which are defined over an exceedingly high-dimensional parameter space. It is then essential to use a fairly large amount of Monte Carlo samples to obtain an approximation with a high-degree of accuracy. In this paper, we address some implementation issues on large-scale particle filtering, and then, indicate the importance of large-scale computing for parameter learning of in silico biological pathways. We have tested the ability of the particle filtering with 10(8) Monte Carlo samples on the transcription circuit of circadian clock that contains 45 unknown kinetic parameters. The proposed approach could reveal clearly the shape of the posterior distributions over the 45 dimensional parameter space.

Bayesian learning of biological pathways on genomic data assimilation

MOTIVATION Mathematical modeling and simulation, based on biochemical rate equations, provide us a rigorous tool for unraveling complex mechanisms of biological pathways. To proceed to simulation experiments, it is an essential first step to find effective values of model parameters, which are difficult to measure from in vivo and in vitro experiments. Furthermore, once a set of hypothetical models has been created, any statistical criterion is needed to test the ability of the constructed models and to proceed to model revision. RESULTS The aim of our research is to present a new statistical technology towards data-driven construction of in silico biological pathways. The method starts with a knowledge-based modeling with hybrid functional Petri net. It then proceeds to the Bayesian learning of model parameters for which experimental data are available. This process exploits quantitative measurements of evolving biochemical reactions, e.g. gene expression data. Another important issue that we consider is statistical evaluation and comparison of the constructed hypothetical pathways. For this purpose, we have developed a new Bayesian information-theoretic measure that assesses the predictability and the biological robustness of in silico pathways. AVAILABILITY The FORTRAN source codes are available at the URL http://daweb.ism.ac.jpyoshidar/GDA/ SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Elevated β-catenin pathway as a novel target for patients with resistance to EGF receptor targeting drugs.

There is a high death rate of lung cancer patients. Epidermal growth factor receptor tyrosine kinase inhibitors (EGFR-TKIs) are effective in some lung adenocarcinoma patients with EGFR mutations. However, a significant number of patients show primary and acquire resistance to EGFR-TKIs. Although the Akt kinase is commonly activated due to various resistance mechanisms, the key targets of Akt remain unclear. Here, we show that the Akt-β-catenin pathway may be a common resistance mechanism. We analyzed gene expression profiles of gefitinib-resistant PC9M2 cells that were derived from gefitinib-sensitive lung cancer PC9 cells and do not have known resistance mechanisms including EGFR mutation T790M. We found increased expression of Axin, a β-catenin target gene, increased phosphorylation of Akt and GSK3, accumulation of β-catenin in the cytoplasm/nucleus in PC9M2 cells. Both knockdown of β-catenin and treatment with a β-catenin inhibitor at least partially restored gefitinib sensitivity to PC9M2 cells. Lung adenocarcinoma tissues derived from gefitinib-resistant patients displayed a tendency to accumulate β-catenin in the cytoplasm. We provide a rationale for combination therapy that includes targeting of the Akt-β-catenin pathway to improve the efficacy of EGFR-TKIs.

Automated detection and tracking of many cells by using 4D live-cell imaging data

Motivation: Automated fluorescence microscopes produce massive amounts of images observing cells, often in four dimensions of space and time. This study addresses two tasks of time-lapse imaging analyses; detection and tracking of the many imaged cells, and it is especially intended for 4D live-cell imaging of neuronal nuclei of Caenorhabditis elegans. The cells of interest appear as slightly deformed ellipsoidal forms. They are densely distributed, and move rapidly in a series of 3D images. Thus, existing tracking methods often fail because more than one tracker will follow the same target or a tracker transits from one to other of different targets during rapid moves. Results: The present method begins by performing the kernel density estimation in order to convert each 3D image into a smooth, continuous function. The cell bodies in the image are assumed to lie in the regions near the multiple local maxima of the density function. The tasks of detecting and tracking the cells are then addressed with two hill-climbing algorithms. The positions of the trackers are initialized by applying the cell-detection method to an image in the first frame. The tracking method keeps attacking them to near the local maxima in each subsequent image. To prevent the tracker from following multiple cells, we use a Markov random field (MRF) to model the spatial and temporal covariation of the cells and to maximize the image forces and the MRF-induced constraint on the trackers. The tracking procedure is demonstrated with dynamic 3D images that each contain >100 neurons of C.elegans. Availability: http://daweb.ism.ac.jp/yoshidalab/crest/ismb2014 Supplementary information: Supplementary data are available at http://daweb.ism.ac.jp/yoshidalab/crest/ismb2014 Contact: yoshidar@ism.ac.jp