Prediction of condition-specific regulatory genes using machine learning.

Recent advances in genomic technologies have generated data on large-scale protein-DNA interactions and open chromatin regions for many eukaryotic species. How to identify condition-specific functions of transcription factors using these data has become a major challenge in genomic research. To solve this problem, we have developed a method called ConSReg, which provides a novel approach to integrate regulatory genomic data into predictive machine learning models of key regulatory genes. Using Arabidopsis as a model system, we tested our approach to identify regulatory genes in data sets from single cell gene expression and from abiotic stress treatments. Our results showed that ConSReg accurately predicted transcription factors that regulate differentially expressed genes with an average auROC of 0.84, which is 23.5-25% better than enrichment-based approaches. To further validate the performance of ConSReg, we analyzed an independent data set related to plant nitrogen responses. ConSReg provided better rankings of the correct transcription factors in 61.7% of cases, which is three times better than other plant tools. We applied ConSReg to Arabidopsis single cell RNA-seq data, successfully identifying candidate regulatory genes that control cell wall formation. Our methods provide a new approach to define candidate regulatory genes using integrated genomic data in plants.

Proteome-wide, Structure-Based Prediction of Protein-Protein Interactions/New Molecular Interactions Viewer.

Determining the complete Arabidopsis (Arabidopsis thaliana) protein-protein interaction network is essential for understanding the functional organization of the proteome. Numerous small-scale studies and a couple of large-scale ones have elucidated a fraction of the estimated 300,000 binary protein-protein interactions in Arabidopsis. In this study, we provide evidence that a docking algorithm has the ability to identify real interactions using both experimentally determined and predicted protein structures. We ranked 0.91 million interactions generated by all possible pairwise combinations of 1,346 predicted structure models from an Arabidopsis predicted "structure-ome" and found a significant enrichment of real interactions for the top-ranking predicted interactions, as shown by cosubcellular enrichment analysis and yeast two-hybrid validation. Our success rate for computationally predicted, structure-based interactions was 63% of the success rate for published interactions naively tested using the yeast two-hybrid system and 2.7 times better than for randomly picked pairs of proteins. This study provides another perspective in interactome exploration and biological network reconstruction using protein structural information. We have made these interactions freely accessible through an improved Arabidopsis Interactions Viewer and have created community tools for accessing these and ∼2.8 million other protein-protein and protein-DNA interactions for hypothesis generation by researchers worldwide. The Arabidopsis Interactions Viewer is freely available at http://bar.utoronto.ca/interactions2/.

Potential targets of VIVIPAROUS1/ABI3-LIKE1 (VAL1) repression in developing Arabidopsis thaliana embryos.

Developing Arabidopsis seeds accumulate oils and seed storage proteins synthesized by the pathways of primary metabolism. Seed development and metabolism are positively regulated by transcription factors belonging to the LAFL (LEC1, AB13, FUSCA3 and LEC2) regulatory network. The VAL gene family encodes repressors of the seed maturation program in germinating seeds, although they are also expressed during seed maturation. The possible regulatory role of VAL1 in seed development has not been studied to date. Reverse genetics revealed that val1 mutant seeds accumulated elevated levels of proteins compared with the wild type, suggesting that VAL1 functions as a repressor of seed metabolism; however, in the absence of VAL1, the levels of metabolites, ABA, auxin and jasmonate derivatives did not change significantly in developing embryos. Two VAL1 splice variants were identified through RNA sequencing analysis: a full-length form and a truncated form lacking the plant homeodomain-like domain associated with epigenetic repression. None of the transcripts encoding the core LAFL network transcription factors were affected in val1 embryos. Instead, activation of VAL1 by FUSCA3 appears to result in the repression of a subset of seed maturation genes downstream of core LAFL regulators, as 39% of transcripts in the FUSCA3 regulon were derepressed in the val1 mutant. The LEC1 and LEC2 regulons also responded, but to a lesser extent. Additional 832 transcripts that were not LAFL targets were derepressed in val1 mutant embryos. These transcripts are candidate targets of VAL1, acting through epigenetic and/or transcriptional repression.

CoSpliceNet: a framework for co-splicing network inference from transcriptomics data.

BACKGROUND: Alternative splicing has been proposed to increase transcript diversity and protein plasticity in eukaryotic organisms, but the extent to which this is the case is currently unclear, especially with regard to the diversification of molecular function. Eukaryotic splicing involves complex interactions of splicing factors and their targets. Inference of co-splicing networks capturing these types of interactions is important for understanding this crucial, highly regulated post-transcriptional process at the systems level. RESULTS: First, several transcript and protein attributes, including coding potential of transcripts and differences in functional domains of proteins, were compared between splice variants and protein isoforms to assess transcript and protein diversity in a biological system. Alternative splicing was shown to increase transcript and function-related protein diversity in developing Arabidopsis embryos. Second, CoSpliceNet, which integrates co-expression and motif discovery at splicing regulatory regions to infer co-splicing networks, was developed. CoSpliceNet was applied to temporal RNA sequencing data to identify candidate regulators of splicing events and predict RNA-binding motifs, some of which are supported by prior experimental evidence. Analysis of inferred splicing factor targets revealed an unexpected role for the unfolded protein response in embryo development. CONCLUSIONS: The methods presented here can be used in any biological system to assess transcript diversity and protein plasticity and to predict candidate regulators, their targets, and RNA-binding motifs for splicing factors. CoSpliceNet is freely available at http://delasa.github.io/co-spliceNet/ .

Transcriptome-wide functional characterization reveals novel relationships among differentially expressed transcripts in developing soybean embryos.

BACKGROUND: Transcriptomics reveals the existence of transcripts of different coding potential and strand orientation. Alternative splicing (AS) can yield proteins with altered number and types of functional domains, suggesting the global occurrence of transcriptional and post-transcriptional events. Many biological processes, including seed maturation and desiccation, are regulated post-transcriptionally (e.g., by AS), leading to the production of more than one coding or noncoding sense transcript from a single locus. RESULTS: We present an integrated computational framework to predict isoform-specific functions of plant transcripts. This framework includes a novel plant-specific weighted support vector machine classifier called CodeWise, which predicts the coding potential of transcripts with over 96 % accuracy, and several other tools enabling global sequence similarity, functional domain, and co-expression network analyses. First, this framework was applied to all detected transcripts (103,106), out of which 13 % was predicted by CodeWise to be noncoding RNAs in developing soybean embryos. Second, to investigate the role of AS during soybean embryo development, a population of 2,938 alternatively spliced and differentially expressed splice variants was analyzed and mined with respect to timing of expression. Conserved domain analyses revealed that AS resulted in global changes in the number, types, and extent of truncation of functional domains in protein variants. Isoform-specific co-expression network analysis using ArrayMining and clustering analyses revealed specific sub-networks and potential interactions among the components of selected signaling pathways related to seed maturation and the acquisition of desiccation tolerance. These signaling pathways involved abscisic acid- and FUSCA3-related transcripts, several of which were classified as noncoding and/or antisense transcripts and were co-expressed with corresponding coding transcripts. Noncoding and antisense transcripts likely play important regulatory roles in seed maturation- and desiccation-related signaling in soybean. CONCLUSIONS: This work demonstrates how our integrated framework can be implemented to make experimentally testable predictions regarding the coding potential, co-expression, co-regulation, and function of transcripts and proteins related to a biological process of interest.

Evidence for extensive heterotrophic metabolism, antioxidant action, and associated regulatory events during winter hardening in Sitka spruce.

BACKGROUND: Cold acclimation in woody perennials is a metabolically intensive process, but coincides with environmental conditions that are not conducive to the generation of energy through photosynthesis. While the negative effects of low temperatures on the photosynthetic apparatus during winter have been well studied, less is known about how this is reflected at the level of gene and metabolite expression, nor how the plant generates primary metabolites needed for adaptive processes during autumn. RESULTS: The MapMan tool revealed enrichment of the expression of genes related to mitochondrial function, antioxidant and associated regulatory activity, while changes in metabolite levels over the time course were consistent with the gene expression patterns observed. Genes related to thylakoid function were down-regulated as expected, with the exception of plastid targeted specific antioxidant gene products such as thylakoid-bound ascorbate peroxidase, components of the reactive oxygen species scavenging cycle, and the plastid terminal oxidase. In contrast, the conventional and alternative mitochondrial electron transport chains, the tricarboxylic acid cycle, and redox-associated proteins providing reactive oxygen species scavenging generated by electron transport chains functioning at low temperatures were all active. CONCLUSIONS: A regulatory mechanism linking thylakoid-bound ascorbate peroxidase action with "chloroplast dormancy" is proposed. Most importantly, the energy and substrates required for the substantial metabolic remodeling that is a hallmark of freezing acclimation could be provided by heterotrophic metabolism.

Effects of drought on gene expression in maize reproductive and leaf meristem tissue revealed by RNA-Seq.

Drought stress affects cereals especially during the reproductive stage. The maize (Zea mays) drought transcriptome was studied using RNA-Seq analysis to compare drought-treated and well-watered fertilized ovary and basal leaf meristem tissue. More drought-responsive genes responded in the ovary compared with the leaf meristem. Gene Ontology enrichment analysis revealed a massive decrease in transcript abundance of cell division and cell cycle genes in the drought-stressed ovary only. Among Gene Ontology categories related to carbohydrate metabolism, changes in starch and Suc metabolism-related genes occurred in the ovary, consistent with a decrease in starch levels, and in Suc transporter function, with no comparable changes occurring in the leaf meristem. Abscisic acid (ABA)-related processes responded positively, but only in the ovaries. Related responses suggested the operation of low glucose sensing in drought-stressed ovaries. The data are discussed in the context of the susceptibility of maize kernel to drought stress leading to embryo abortion and the relative robustness of dividing vegetative tissue taken at the same time from the same plant subjected to the same conditions. Our working hypothesis involves signaling events associated with increased ABA levels, decreased glucose levels, disruption of ABA/sugar signaling, activation of programmed cell death/senescence through repression of a phospholipase C-mediated signaling pathway, and arrest of the cell cycle in the stressed ovary at 1 d after pollination. Increased invertase levels in the stressed leaf meristem, on the other hand, resulted in that tissue maintaining hexose levels at an "unstressed" level, and at lower ABA levels, which was correlated with successful resistance to drought stress.

Comparing time series transcriptome data between plants using a network module finding algorithm.

BACKGROUND: Comparative transcriptome analysis is the comparison of expression patterns between homologous genes in different species. Since most molecular mechanistic studies in plants have been performed in model species, including Arabidopsis and rice, comparative transcriptome analysis is particularly important for functional annotation of genes in diverse plant species. Many biological processes, such as embryo development, are highly conserved between different plant species. The challenge is to establish one-to-one mapping of the developmental stages between two species. RESULTS: In this manuscript, we solve this problem by converting the gene expression patterns into co-expression networks and then apply network module finding algorithms to the cross-species co-expression network. We describe how such analyses are carried out using bash scripts for preliminary data processing followed by using the R programming language for module finding with a simulated annealing method. We also provide instructions on how to visualize the resulting co-expression networks across species. CONCLUSIONS: We provide a comprehensive pipeline from installing software and downloading raw transcriptome data to predicting homologous genes and finding orthologous co-expression networks. From the example provided, we demonstrate the application of our method to reveal functional conservation and divergence of genes in two plant species.

Physiological and molecular adaptations to drought in Andean potato genotypes.

The drought stress tolerance of two Solanum tuberosum subsp. andigena landraces, one hybrid (adgxtbr) and Atlantic (S. tuberosum subsp. tuberosum) has been evaluated. Photosynthesis in the Andigena landraces during prolonged drought was maintained significantly longer than in the Tuberosum (Atlantic) line. Among the Andigena landraces, 'Sullu' (SUL) was more drought resistant than 'Negra Ojosa' (NOJ). Microarray analysis and metabolite data from leaf samples taken at the point of maximum stress suggested higher mitochondrial metabolic activity in SUL than in NOJ. A greater induction of chloroplast-localized antioxidant and chaperone genes in SUL compared with NOJ was evident. ABA-responsive TFs were more induced in NOJ compared with SUL, including WRKY1, mediating a response in SA signalling that may give rise to increased ROS. NOJ may be experiencing higher ROS levels than SUL. Metabolite profiles of NOJ were characterized by compounds indicative of stress, for example, proline, trehalose, and GABA, which accumulated to a higher degree than in SUL. The differences between the Andigena lines were not explained by protective roles of compatible solutes; hexoses and complex sugars were similar in both landraces. Instead, lower levels of ROS accumulation, greater mitochondrial activity and active chloroplast defences contributed to a lower stress load in SUL than in NOJ during drought.

Tuber development phenotypes in adapted and acclimated, drought-stressed Solanum tuberosum ssp. andigena have distinct expression profiles of genes associated with carbon metabolism.

A drought screen identified accessions of Solanum tuberosum ssp. andigena that showed varying degrees of physiological acclimation or adaptation to repeated drought stress. The accessions also showed variable tuber phenotypes from small tubers that failed to develop in an accession that showed photosynthetic adaptation to normal tubers in an accession with a phenotype showing some degree of photosynthetic adaptation and acclimation. Using microarray data, we correlated the expression of genes associated with carbon metabolism with the tuber development phenotypes under drought. Genes associated with sucrose and starch metabolism showed responses consistent with starch deficiency in the adapted accession and normal starch deposition in the intermediate accession. Starch phosphorylase and glycogen bound starch synthase were induced in the adapted accession, which had abnormal tuber development. Genes associated with trehalose were induced in the intermediate accession with normal tuber development. Genes associated with respiration were also induced in the intermediate accession, and a pattern compatible with the existence of a 3PGA recovery pathway was revealed. Expression of thioredoxin genes also correlated with tuber development phenotypes under drought stress. The data suggest differential regulation of starch deposition in accessions of Andigena with different abilities to respond to drought stress.

Identification of regulatory modules in genome scale transcription regulatory networks.

BACKGROUND: In gene regulatory networks, transcription factors often function as co-regulators to synergistically induce or inhibit expression of their target genes. However, most existing module-finding algorithms can only identify densely connected genes but not co-regulators in regulatory networks. METHODS: We have developed a new computational method, CoReg, to identify transcription co-regulators in large-scale regulatory networks. CoReg calculates gene similarities based on number of common neighbors of any two genes. Using simulated and real networks, we compared the performance of different similarity indices and existing module-finding algorithms and we found CoReg outperforms other published methods in identifying co-regulatory genes. We applied CoReg to a large-scale network of Arabidopsis with more than 2.8 million edges and we analyzed more than 2,300 published gene expression profiles to charaterize co-expression patterns of gene moduled identified by CoReg. RESULTS: We identified three types of modules in the Arabidopsis network: regulator modules, target modules and intermediate modules. Regulator modules include genes with more than 90% edges as out-going edges; Target modules include genes with more than 90% edges as incoming edges. Other modules are classified as intermediate modules. We found that genes in target modules tend to be highly co-expressed under abiotic stress conditions, suggesting this network struture is robust against perturbation. CONCLUSIONS: Our analysis shows that the CoReg is an accurate method in identifying co-regulatory genes in large-scale networks. We provide CoReg as an R package, which can be applied in finding co-regulators in any organisms with genome-scale regulatory network data.

Beacon Editor: Capturing Signal Transduction Pathways Using the Systems Biology Graphical Notation Activity Flow Language.

The Beacon Editor is a cross-platform desktop application for the creation and modification of signal transduction pathways using the Systems Biology Graphical Notation Activity Flow (SBGN-AF) language. Prompted by biologists' requests for enhancements, the Beacon Editor includes numerous powerful features for the benefit of creation and presentation.

Expresso: A database and web server for exploring the interaction of transcription factors and their target genes in Arabidopsis thaliana using ChIP-Seq peak data.

Motivation: The increasing availability of chromatin immunoprecipitation sequencing (ChIP-Seq) data enables us to learn more about the action of transcription factors in the regulation of gene expression. Even though in vivo transcriptional regulation often involves the concerted action of more than one transcription factor, the format of each individual ChIP-Seq dataset usually represents the action of a single transcription factor. Therefore, a relational database in which available ChIP-Seq datasets are curated is essential. Results: We present Expresso (database and webserver) as a tool for the collection and integration of available Arabidopsis ChIP-Seq peak data, which in turn can be linked to a user's gene expression data. Known target genes of transcription factors were identified by motif analysis of publicly available GEO ChIP-Seq data sets. Expresso currently provides three services: 1) Identification of target genes of a given transcription factor; 2) Identification of transcription factors that regulate a gene of interest; 3) Computation of correlation between the gene expression of transcription factors and their target genes. Availability: Expresso is freely available at http://bioinformatics.cs.vt.edu/expresso/.

Multilevel Regulation of Abiotic Stress Responses in Plants.

The sessile lifestyle of plants requires them to cope with stresses in situ. Plants overcome abiotic stresses by altering structure/morphology, and in some extreme conditions, by compressing the life cycle to survive the stresses in the form of seeds. Genetic and molecular studies have uncovered complex regulatory processes that coordinate stress adaptation and tolerance in plants, which are integrated at various levels. Investigating natural variation in stress responses has provided important insights into the evolutionary processes that shape the integrated regulation of adaptation and tolerance. This review primarily focuses on the current understanding of how transcriptional, post-transcriptional, post-translational, and epigenetic processes along with genetic variation orchestrate stress responses in plants. We also discuss the current and future development of computational tools to identify biologically meaningful factors from high dimensional, genome-scale data and construct the signaling networks consisting of these components.

XcisClique: analysis of regulatory bicliques.

BACKGROUND: Modeling of cis-elements or regulatory motifs in promoter (upstream) regions of genes is a challenging computational problem. In this work, set of regulatory motifs simultaneously present in the promoters of a set of genes is modeled as a biclique in a suitably defined bipartite graph. A biologically meaningful co-occurrence of multiple cis-elements in a gene promoter is assessed by the combined analysis of genomic and gene expression data. Greater statistical significance is associated with a set of genes that shares a common set of regulatory motifs, while simultaneously exhibiting highly correlated gene expression under given experimental conditions. METHODS: XcisClique, the system developed in this work, is a comprehensive infrastructure that associates annotated genome and gene expression data, models known cis-elements as regular expressions, identifies maximal bicliques in a bipartite gene-motif graph; and ranks bicliques based on their computed statistical significance. Significance is a function of the probability of occurrence of those motifs in a biclique (a hypergeometric distribution), and on the new sum of absolute values statistic (SAV) that uses Spearman correlations of gene expression vectors. SAV is a statistic well-suited for this purpose as described in the discussion. RESULTS: XcisClique identifies new motif and gene combinations that might indicate as yet unidentified involvement of sets of genes in biological functions and processes. It currently supports Arabidopsis thaliana and can be adapted to other organisms, assuming the existence of annotated genomic sequences, suitable gene expression data, and identified regulatory motifs. A subset of Xcis Clique functionalities, including the motif visualization component MotifSee, source code, and supplementary material are available at https://bioinformatics.cs.vt.edu/xcisclique/.

The statistics of identifying differentially expressed genes in Expresso and TM4: a comparison.

BACKGROUND: Analysis of DNA microarray data takes as input spot intensity measurements from scanner software and returns differential expression of genes between two conditions, together with a statistical significance assessment. This process typically consists of two steps: data normalization and identification of differentially expressed genes through statistical analysis. The Expresso microarray experiment management system implements these steps with a two-stage, log-linear ANOVA mixed model technique, tailored to individual experimental designs. The complement of tools in TM4, on the other hand, is based on a number of preset design choices that limit its flexibility. In the TM4 microarray analysis suite, normalization, filter, and analysis methods form an analysis pipeline. TM4 computes integrated intensity values (IIV) from the average intensities and spot pixel counts returned by the scanner software as input to its normalization steps. By contrast, Expresso can use either IIV data or median intensity values (MIV). Here, we compare Expresso and TM4 analysis of two experiments and assess the results against qRT-PCR data. RESULTS: The Expresso analysis using MIV data consistently identifies more genes as differentially expressed, when compared to Expresso analysis with IIV data. The typical TM4 normalization and filtering pipeline corrects systematic intensity-specific bias on a per microarray basis. Subsequent statistical analysis with Expresso or a TM4 t-test can effectively identify differentially expressed genes. The best agreement with qRT-PCR data is obtained through the use of Expresso analysis and MIV data. CONCLUSION: The results of this research are of practical value to biologists who analyze microarray data sets. The TM4 normalization and filtering pipeline corrects microarray-specific systematic bias and complements the normalization stage in Expresso analysis. The results of Expresso using MIV data have the best agreement with qRT-PCR results. In one experiment, MIV is a better choice than IIV as input to data normalization and statistical analysis methods, as it yields as greater number of statistically significant differentially expressed genes; TM4 does not support the choice of MIV input data. Overall, the more flexible and extensive statistical models of Expresso achieve more accurate analytical results, when judged by the yardstick of qRT-PCR data, in the context of an experimental design of modest complexity.

A Machine Learning Approach to Predict Gene Regulatory Networks in Seed Development in Arabidopsis.

Gene regulatory networks (GRNs) provide a representation of relationships between regulators and their target genes. Several methods for GRN inference, both unsupervised and supervised, have been developed to date. Because regulatory relationships consistently reprogram in diverse tissues or under different conditions, GRNs inferred without specific biological contexts are of limited applicability. In this report, a machine learning approach is presented to predict GRNs specific to developing Arabidopsis thaliana embryos. We developed the Beacon GRN inference tool to predict GRNs occurring during seed development in Arabidopsis based on a support vector machine (SVM) model. We developed both global and local inference models and compared their performance, demonstrating that local models are generally superior for our application. Using both the expression levels of the genes expressed in developing embryos and prior known regulatory relationships, GRNs were predicted for specific embryonic developmental stages. The targets that are strongly positively correlated with their regulators are mostly expressed at the beginning of seed development. Potential direct targets were identified based on a match between the promoter regions of these inferred targets and the cis elements recognized by specific regulators. Our analysis also provides evidence for previously unknown inhibitory effects of three positive regulators of gene expression. The Beacon GRN inference tool provides a valuable model system for context-specific GRN inference and is freely available at https://github.com/BeaconProjectAtVirginiaTech/beacon_network_inference.git.

A rootstock provides water conservation for a grafted commercial tomato (Solanum lycopersicum L.) line in response to mild-drought conditions: a focus on vegetative growth and photosynthetic parameters.

The development of water stress resistant lines of commercial tomato by breeding or genetic engineering is possible, but will take considerable time before commercial varieties are available for production. However, grafting commercial tomato lines on drought resistant rootstock may produce drought tolerant commercial tomato lines much more rapidly. Due to changing climates and the need for commercial production of vegetables in low quality fields there is an urgent need for stress tolerant commercial lines of vegetables such as tomato. In previous observations we identified a scion root stock combination ('BHN 602' scion grafted onto 'Jjak Kkung' rootstock hereafter identified as 602/Jjak) that had a qualitative drought-tolerance phenotype when compared to the non-grafted line. Based on this initial observation, we studied photosynthesis and vegetative above-ground growth during mild-drought for the 602/Jjak compared with another scion-rootstock combination ('BHN 602' scion grafted onto 'Cheong Gang' rootstock hereafter identified as 602/Cheong) and a non-grafted control. Overall above ground vegetative growth was significantly lower for 602/Jjak in comparison to the other plant lines. Moreover, water potential reduction in response to mild drought was significantly less for 602/Jjak, yet stomatal conductance of all plant-lines were equally inhibited by mild-drought. Light saturated photosynthesis of 602/Jjak was less affected by low water potential than the other two lines as was the % reduction in mesophyll conductance. Therefore, the Jjak Kkung rootstock caused aboveground growth reduction, water conservation and increased photosynthetic tolerance of mild drought. These data show that different rootstocks can change the photosynthetic responses to drought of a high yielding, commercial tomato line. Also, this rapid discovery of one scion-rootstock combination that provided mild-drought tolerance suggests that screening more scion-rootstock combination for stress tolerance may rapidly yield commercially viable, stress tolerant lines of tomato.