- Open Access
Genome-wide prediction and analysis of human tissue-selective genes using microarray expression data
© Teng et al.; licensee BioMed Central Ltd. 2013
- Published: 23 January 2013
Understanding how genes are expressed specifically in particular tissues is a fundamental question in developmental biology. Many tissue-specific genes are involved in the pathogenesis of complex human diseases. However, experimental identification of tissue-specific genes is time consuming and difficult. The accurate predictions of tissue-specific gene targets could provide useful information for biomarker development and drug target identification.
In this study, we have developed a machine learning approach for predicting the human tissue-specific genes using microarray expression data. The lists of known tissue-specific genes for different tissues were collected from UniProt database, and the expression data retrieved from the previously compiled dataset according to the lists were used for input vector encoding. Random Forests (RFs) and Support Vector Machines (SVMs) were used to construct accurate classifiers. The RF classifiers were found to outperform SVM models for tissue-specific gene prediction. The results suggest that the candidate genes for brain or liver specific expression can provide valuable information for further experimental studies. Our approach was also applied for identifying tissue-selective gene targets for different types of tissues.
A machine learning approach has been developed for accurately identifying the candidate genes for tissue specific/selective expression. The approach provides an efficient way to select some interesting genes for developing new biomedical markers and improve our knowledge of tissue-specific expression.
- Support Vector Machine
- Random Forest
- Gene Prediction
- Matthews Correlation Coefficient
- Random Forest Classifier
Understanding how different tissues achieve specificity is a fundamental question in tissue ontogenesis and evolution. Some genes are highly expressed in a particular tissue and lowly expressed or not expressed in other tissues. These genes are generally called tissue-selective genes. The genes are responsible for specialized functions in particular tissues, and thus can serve as the biomarkers for specific biological processes. In addition, many tissue-selective genes are involved in the pathogenesis of complex human diseases , including insulin signaling pathways in diabetes  and tumor-host interactions in cancer . Since the majority of disease genes have the tendency to be expressed preferentially in particular tissues , identifying tissue-selective genes is also important for drug target selection in biomedical research. Tissue-specific genes, which are specifically expressed in a particular tissue, are regarded as the special case of tissue selective genes. The identification of tissue-specific genes could help biologists to elucidate the molecular mechanisms of tissue development and provide valuable information for identifying candidate biomarkers and drug targets.
Different methods have been used to identify and characterize tissue-specific genes. Traditional experimental methods, including RT-PCR and Northern blot, are usually carried out at the single-gene level and thus time-consuming. High-throughput technologies, such as Expressed Sequence Tag (EST) sequencing and DNA microarrays, have the capacity to perform genome-wide analysis with high efficiency. The DNA microarray technology can generate large amounts of gene expression data from various tissues, and provide the useful data source for analyzing tissue-specific genes. Several statistical methods have been applied for identifying tissue-specific genes using gene expression data. Kadota and co-workers  described an unsupervised method to select the tissue-specific genes using Akaike's information criterion (AIC) approach. Another method called ROKU  has been developed by the same group for detecting tissue-specific gene expression patterns. The approach used Shannon entropy and outlier detection to scan expression profiles for ranking tissue-specific genes. Liang et al.  developed a statistical method based on hypothesis testing procedures to profile and identify the tissue-selective genes. However, the statistical methods for tissue-specific gene prediction suffer from drawbacks. The microarray expression data are generated from different experiments, both biological variations and experimental noise result in significant variations in data quality. The statistical methods usually assigned an equal weight to each observation for prediction. Thus, the methods do not work well for non-linear models and may not detect the hidden expression patterns from the noisy microarray data. Moreover, the statistical methods do not use biological knowledge for prediction. The simple data-driven analysis may produce some misleading results for further experimental studies.
Machine learning can automatically recognize hidden patterns in complex data. It has been shown that machine learning can be used to construct accurate classifiers for tissue-specific gene prediction. Chikina et al.  used Support Vector Machines (SVMs) to predict tissue-specific gene expression in Caenorhabditis elegans with whole-animal microarray data. The SVM classifiers reached high predictive performances in nearly all tissues. It was shown that the approach outperformed clustering methods and provided valuable information for further experimental studies. However, it is still unknown whether machine learning methods can be used to predict tissue-specific genes in human.
We previously compiled a large microarray gene expression dataset, which contained 2,968 expression profiles of various human tissues, including brain, liver, testis, blood and kidney samples . A computational method was also developed for identifying tissue-selective genes using the integrated microarray dataset. However, the method assigned an equal weight to each expression profile for identification. In this study, a machine learning approach was developed for human tissue-specific/selective gene prediction using the available dataset. According to the lists of known tissue-specific/selective genes, the gene expression data were extracted from the compiled dataset and used for classifier construction. Random Forests (RFs) and Support Vector Machines (SVMs) were trained with the expression data to construct accurate classifiers. The results indicate that the RF classifiers achieved better predictive performance for tissue-specific gene prediction. The approach generated large numbers of candidate genes for brain and liver-specific expression. The examinations of high scoring genes suggest that our approach can be used to select candidate genes for experimental studies.
Microarray data collection and integration
The approach for compiling human microarray expression profiles was described in our previous study . The microarray gene expression profiles from 131 microarray studies (different experimental batches) were collected from the NCBI GEO database. These expression profiles were generated using the Affymetrix HG-U133 Plus 2.0 Array with 54,613 probe sets. The statistical model-based method, dChip , was used for microarray data normalization. The raw data in CEL file format were divided into different normalization groups. The invariant set method  was used to normalize each group of microarray profiles to minimize the batch effect (variation across different microarray studies). After normalization, global median transformation was used to integrate the microarray profiles into a single dataset. The dataset used in this study contained 2,968 expression profiles of various human tissues, including brain (616 profiles), liver (117 profiles), testis (36 profiles), blood (409 profiles) and kidney (73 profiles).
Training data preparation
Tissue-selective genes are defined as the genes whose expression is enriched for one or a few similar tissue types. The genes were manually collected from the UniProt database. The particular tissue name was used as a query and the reviewed human genes were selected for preparation. The tissue-selective genes are defined as the genes that are preferentially expressed in a particular tissue from the descriptions of their annotations. Most of the genes are identified by the experimental methods, which are independent from the microarray expression data in the list. In this study, 408 brain-selective genes, 96 liver-selective genes, 326 testis-selective genes, 324 blood-selective genes and 45 kidney-selective genes were collected from UniProt database. Tissue-specific genes, whose expression is specific to only one particular tissue type, are considered as the special case of tissue-selective genes. 289 brain-specific genes and 69 liver-specific genes were selected from the corresponding tissue-selective genes with the annotation that their expression is specific to only brain or liver. Tissue specific-genes can be used to construct the highly accurate classifiers. Thus, the tissue specific-gene prediction is the focus of the present study.
According to the known tissue-specific/selective gene lists, the tissue-specific/selective gene expression data was retrieved from the integrated microarray dataset and labelled as the positive training instances. The probe sets with detectable expression signals in corresponding tissue samples were selected for classifier construction. For tissue-specific gene prediction, the expression values for 403 probe sets of brain-specific genes and 90 probe sets of liver-specific genes were used for input vector encoding. 692 probe sets of brain-selective genes, 150 probe sets of liver-selective, 430 probe sets of testis-selective genes, 456 probe sets of blood-selective genes and 76 probe sets of kidney-selective genes were used for tissue-selective gene prediction.
The negative examples were defined as the genes that do not have preferential expression in particular tissues. For this study, we randomly selected the data instances from the remaining data and labelled as the negative training instances. The number of negative instances was set as tenfold with positive instances to make enough data instances for training. The negative and positive data instances were combined as the training dataset to construct classifiers using machine learning algorithms. The remaining probes were used as the candidate genes for prediction with the classifiers constructed from the training dataset.
The use of 2,968 expression profiles for input vector encoding gives the same number of input variables. One potential problem is model overfitting since there were only a small number of positive instances (probe sets of known tissue-specific genes) available for this study. We thus used the Random Forest (RF) learning algorithm, which could handle a large number of input variables and avoid model overfitting through random feature selection. The randomForest package in R  was used in this study for classifier construction. The number of trees in a classifier (ntree) and the number of variables selected to split each node (mtry) were set to 1000 and 6, respectively. The classifier performance did not show significant improvement by using other values for the mtry and ntree parameters.
Support Vector Machines
Support Vector Machines (SVMs) are widely used for binary classification . In this study, SVM classifiers were constructed and compared with RF classifiers for identifying human tissue-specific genes. The SVMlight software package (http://svmlight.joachims.org/) was used to construct the SVM classifiers with the linear kernel function . The polynomial and radial basis function (RBF) kernels were also tested for classifier construction, but the classifiers did not achieve high predictive performances in cross-validation tests.
Classifier evaluation and prediction
where TP is the number of true positives; TN is the number of true negatives; FP is the number of false positives; and FN is the number of false negatives. The Receiver Operating Characteristic (ROC) curve  and the area under the curve (AUC)  were also used for classifier evaluation and comparison.
The classifier construction and prediction were repeated ten times. In each run, classifier performance was evaluated using the above measures. The classifier was then used to predict tissue-specific genes in the human genome. The tissue-specific gene targets were sorted according to the average value of classifier outputs from ten predictions, and a higher value might indicate a higher probability of being expressed predominantly in a particular tissue.
Go enrichment and promoter sequence analyses
DAVID (http://david.abcc.ncifcrf.gov/), a web-based tool for go enrichment analysis, was used for integrating functional annotations of predicted tissue-specific genes . The probe names of candidate targets were used as the inputs, and "GOTERM_BP_FAT", "GOTERM_CC_FAT" and "GOTERM_MF_FAT" were utilized as the Gene Ontology search option.
The 500 bp upstream sequences of predicted tissue-specific genes were downloaded from the UCSC Table Browser (http://genome.ucsc.edu/cgi-bin/hgTables) with the gene names of candidate targets as the inputs. The promoter sequences were used to identify the regulatory DNA motifs with SCOPE web server (http://genie.dartmouth.edu/scope/). SCOPE  applied three learning algorithms to find different type of sequence motifs, including non-degenerate motifs, degenerate motifs, and bipartite motifs. The candidate regulatory motifs used in this study are the high scoring motifs (motif width ≥ 6) returned from the combination of the prediction results. DiRE  (http://dire.dcode.org/) were used to analyse the candidate transcription factors of predicted tissue-specific genes. The gene names of candidate targets were used as the inputs, and the transcription factors for the regulatory elements of candidate targets identified with the Enhancer Identification method were reported.
The transcripts of known liver-specific genes are detected clearly in liver samples (Figure 2). The results suggest that the expression data according to our lists of known tissue-specific genes can provide useful information for classifier construction using machine learning methods. It is noteworthy that some probe sets of known tissue-specific genes have high expression or no expression for all tissue samples. To improve the quality of classifiers, the probes without detectable expression signals in all the samples are excluded from the training dataset.
Prediction of tissue-specific genes
Comparison of Random Forest and Support Vector Machine classifiers for predicting tissue-specific genes.
95.82 (± 0.263)
0.5091 (± 0.015)
97.43 (± 0.153)
0.5676 (± 0.016)
0.9488 (± 0.002)
98.61 (± 0.309)
0.8350 (± 0.025)
0.9854 (± 0.004)
99.12 (± 0.255)
0.8290 (± 0.0213)
0.9777 (± 0.002)
List of high-scoring genes with specific expression in the brain.
Bruno-like 4, RNA binding protein (Drosophila)
Ankyrin repeat and sterile alpha motif domain containing 1B
Tripartite motif-containing 9
Bruno-like 4, RNA binding protein (Drosophila)
Guanine nucleotide binding protein (G protein), alpha activating activity polypeptide O
Heparan sulfate 6-O-sulfotransferase 3
Fas apoptotic inhibitory molecule 2
Similar to RIKEN cDNA D630023F18
Sodium channel, voltage-gated, type II, alpha subunit
Synaptic vesicle glycoprotein 2A
CDNA FLJ37672 fis, clone BRHIP2012059
unc-13 homolog A (C. elegans)
Fas apoptotic inhibitory molecule 2
Neurotrophic tyrosine kinase, receptor, type 2
Opioid binding protein/cell adhesion molecule-like
Clone 24626 mRNA sequence
FXYD domain containing ion transport regulator 7
List of high-scoring genes with specific expression in the liver.
Coagulation factor XI (plasma thromboplastin antecedent)
Serpin peptidase inhibitor, clade C member 1
Aldo-keto reductase family 1, member D1
Coagulation factor IX
Complement component 6
Inter-alpha (globulin) inhibitor H2
Acyl-CoA synthetase medium-chain family member 2
Carboxypeptidase B2 (plasma)
Acyl-CoA synthetase medium-chain family member 2
Nuclear receptor subfamily 0, group B, member 2
Cytochrome P450, family 4, subfamily A, polypeptide 11
Peptidoglycan recognition protein 2
Cytochrome P450, family 4, subfamily A, polypeptide 11
Solute carrier family 22, member 7
GO term enrichment analysis of predicted brain-specific genes.
GO:0019226~transmission of nerve impulse
GO:0005216~ion channel activity
GO:0022838~substrate specific channel activity
GO:0022836~gated channel activity
GO:0022803~passive transmembrane transporter activity
GO term enrichment analysis of predicted liver-specific genes.
GO:0002526~acute inflammatory response
GO:0009611~response to wounding
GO:0002541~activation of plasma proteins involved in acute inflammatory response
In this study, we constructed both RF and SVM classifiers for predicting brain and liver-specific genes. 403 probe sets of brain-specific genes and 90 probe sets of liver-specific genes were used for classifier construction. For brain-specific gene prediction, the RF classifier achieved the AUC value at 0.9488 (Table 1), which is significantly higher than the AUC value of SVM classifier (AUC = 0.8937). The RF classifier reached 53.73% sensitivity and 97.43% specificity, and MCC = 0.5676. For liver-specific gene prediction, the SVM classifier gave MCC = 0.8350 and ROC AUC = 0.9854. The RF classifier achieved a similar level of performance with MCC = 0.8290 and ROC AUC = 0.9777. Thus, the results suggest that the RF algorithm performs better for predicting tissue-specific genes in this study.
The ROC curves of RF and SVM classifiers for predicting brain-specific genes and live-specific genes have been compared in Figure 2. The ROC curves of RF and SVM classifiers are not significantly different for the prediction of liver-specific genes (Figure 2b). However, The ROC curve of RF classifier was clearly better than the SVM classifier for the prediction of brain-specific genes (Figure 2a). The results confirm that RF classifier outperforms the SVM models for tissue-specific gene prediction. The possible reason is that RFs can handle a large number of input variables and avoid model overfitting. The use of 2,968 expression profiles for input vector encoding results in the same large number of input variables, which may lead to model overfitting. Interestingly, the RF algorithm can handle the situation and show better predictive performance in the present study.
Brain-specific gene expression
The human brain gives us the ability to think and sets us apart from other animals. It has a highly complex structure which contains different regions with specific functions. For example, the hippocampus is involved in spatial navigation and long-term memories, whereas the cerebral cortex plays key roles in language, attention and consciousness. Any damage in these regions results in various mental disorders including Alzheimer's disease, Parkinson's disease and Mood disorder. In this study, the predicted brain-specific genes are expected to have preferential expression in the brain, and may play important roles in neuron functions such as synaptic transmission and neuronal migration.
In the study, 1,408 candidate targets with positive scores (the average value of classifier outputs from ten predictions) were predicted as the brain-specific genes (Additional file 1). In Figure 4, the expression patterns of candidate gene targets using RF classifier are visualized with the heat maps generated using TM4 MeV. The predicted targets show clear expression in brain samples, which indicates that our approach is useful for brain-specific gene prediction. Similar to the known brain-specific genes, the transcripts of the predicted targets are also detected in retina samples. GO enrichment analysis of the candidate targets demonstrates that many candidate genes have basic neuron functions (Table 4). For example, neurotransmission is an electrical or chemical signal motion within synapses caused by transmission of a nerve impulse. The predictions are enriched for neurotransmission-related GO terms such as "synapse", "synapse part", "transmission of nerve impulse", "neuron projection", "synaptic transmission" and "passive transmembrane transporter activity". Some channel-related GO terms including "ion channel activity", "substrate specific channel activity", "gated channel activity" and "channel activity" are detected in the enrichment analysis of our predictions.
To understand the regulatory signals of human tissue-specific genes, SCOPE  was used to search for the overrepresented DNA motifs in the promoters of the predicted brain-specific genes (Figure 5a). Some transcription factors might bind to the DNA motifs in the promoter regions and regulate the tissue-specific gene expression. DiRE  was used to identify the candidate transcription factors of the predicted brain-specific genes (Figure 5b). Interestingly, the identified transcription factors might be involved in regulating neuron development and function. For example, BRN2 and OCT belong to the POU domain protein family, and are involved in the differentiation and function of neural cells [21, 22]. X-box-binding protein 1 (XBP1) is activated in neural development and plays a key role for the unfolded protein response in the endoplasmic reticulum . Regulatory factor X1 (RFX1) can enhance the expression of excitatory amino acid transporters (EAAT) to regulate the glutamate neurotransmission . The p53 protein is important for various cellular processes, including apoptosis, differentiation, DNA repair and cell-cycle arrest, in neurons . The results suggest that the approach developed in the present study can provide valuable information for studying the regulatory modules of tissue-specific genes.
The predicted targets have not been annotated as brain-specific genes in the UniProt database. However, recent studies suggest that some of the predicted targets, including BRUNOL4, ANKS1B, TRIM9, NCAN, FAIM2, OPCML and FXYD7 (Table 2), are expressed predominantly in the brain. For example, the RNA-binding protein encoded by BRUNOL4 plays an important role in many cellular processes including RNA stability, pre-mRNA alternative splicing, mRNA editing and translation [26, 27]. It was shown that the protein was predominantly expressed in the brain with enrichment in the hippocampus . In this study, the probes of BRUNOL4 have the highest (223654_s_at, 0.8753) and fourth-ranked (238966_at, 0.8600) scores. ANKS1B encodes Amyloid-beta protein which can regulate the nucleoplasmic coilin protein interactions in neuronal cells. Previous studies showed that the protein is mainly expressed in brain and may be implicated in Alzheimer's disease . Brain-specific E3 ligase encoded by TRIM9 has a high level of expression in the cerebral cortex and may be involved in the pathogenesis of Parkinson's disease . Neurocan (NCAN) modulates neuronal adhesion and migration and is expressed preferentially in the brain . The protein encoded by FAIM2 could protect cells from Fas-mediated apoptosis and shows a high level of expression in the hippocampus . It was shown that OPCML was expressed predominantly in cerebellum and cerebral cortex , whereas FXYD7 was expressed preferentially in the brain .
The other predicted targets were not previously shown to have brain-specific expression. However, some of these genes, including GNAO1, SV2A, SYN2, UNC13A and NTRK2, are involved in basic neuron functions (Table 2). Guanine nucleotide binding protein (GNAO1) mediates the physiological effects of various neuronal receptors . SV2A, SYN2 and UNC13A encode proteins which are important for synaptic transmission in the central and peripheral nervous system [36, 37]. NTRK2 encodes a neurotrophic tyrosine kinase receptor for brain-derived neurotrophic factor (BDNF) and is implicated in childhood mood disorder . By contrast, the functions of some high scoring genes in brain remain to be characterized. HS6ST3 encodes a Heparan sulphate sulfotransferase which plays a key role in the modulation of fibroblast growth factor signalling . The protein encoded by SCN2A forms a voltage-dependent sodium channel and is associated with generalized epilepsy with febrile seizures plus . The corresponding genes of three cDNA sequences (LOC389073, AA879409 and AI186173) were not determined, and their functions in the brain are not clear. The results suggest that the machine learning approach developed in the present study can be used to identify some interesting targets for further experimental studies.
Liver-specific gene expression
The liver is a vital organ for human metabolism, and plays key roles in detoxification, plasma protein synthesis, glycogen storage and hormone production. For example, liver is the source and target organ of inflammatory mediators in the pathogenesis of inflammatory response syndrome , and it is responsible for the production of coagulation factors. Thus, the liver-specific targets identified in this study might be involved in basic liver functions. We identified 493 liver-specific gene targets with positive scores in the analysis (Additional file 2). The functional analysis of the liver-specific gene targets using RF classifier confirms that many of the predicted targets are enriched for liver-related GO terms (Table 5). For example, the GO terms for inflammatory response contained "acute inflammatory response", "response to wounding" and "activation of plasma proteins involved in acute inflammatory response"; the coagulation-related GO terms included "blood coagulation", "coagulation" and "hemostasis". The expression patterns of the predicted liver-specific genes are visualized with the heat map (Figure 3). Clearly, the transcripts of the predicted targets are predominantly detected in liver samples. Some DNA motifs overrepresented in the promoter regions of predicted liver-specific genes have been identified (Figure 5a), and the candidate transcription factors may be involved in the regulation of the specialized functions of liver (Figure 5b). For example, Chicken ovalbumin upstream promoter (COUP) can reduce the expression of hepatocyte nuclear factor 4 (HNF4), which plays a key role in blood coagulation and hepatic metabolism .
As listed in Table 3, 17 of the top 20 high-scoring genes are involved in the metabolism of human liver. The genes, including F11, F9, SERPINC1, APOA2, AKR1D1, ACSM2, ITIH2, PON1, CPB2, AFM, NR0B2, ALB, CYP4A11, PGLYRP2 and SLC22A7, have not been annotated as liver-specific genes in UniProt. However, recent studies suggest that some of these genes are expressed preferentially in the liver. For example, F11, F9 and SERPINC1 are involved in the regulation of blood coagulation cascade . APOA2 encodes apolipoprotein which is synthesized mainly in liver and involved in the metabolism of high density lipoprotein . AKR1D1 encodes the aldo-keto reductase catalyzing the reduction of steroid hormones , whereas ACSM2 encodes enzyme catalyzing the activation of medium-chain length fatty acids . The expression and functions of other three predictions (BG398937, C6 and LPA) have not been documented in the literature.
Tissue-selective gene prediction
Random Forest classifiers for predicting tissue-selective genes.
92.70 (± 0.273)
43.55 (± 1.212)
97.60 (± 0.211)
70.58 (± 0.675)
0.9178 (± 0.002)
96.02 (± 0.341)
65.6 (± 2.499)
99.07 (± 0.191)
82.33 (± 1.293)
0.95467 (± 0.003)
91.00 (± 0.033)
1.49 (± 0.405)
99.95 (± 0.038)
50.72 (± 0.193)
0.8433 (± 0.004)
93.29 (± 0.190)
40.20 (± 1.291)
98.53 (± 0.108)
69.37 (± 0.677)
0.9170 (± 0.002)
93.62 (± 0.508)
26.43 (± 5.355)
99.73 (± 0.159)
63.08 (± 2.703)
0.9300 (± 0.003)
The RF classifiers gave high predictive performance for predicting genes that have preferential expression in other tissue types. The testis is the male sex gland, which produces sperm, male reproductive cell and sex hormones. The classifier for testis-selective gene prediction reached predictive performance with overall accuracy at 91.00% and ROC AUC = 0.8433. The blood transports oxygen and nutrients to other tissues and carries away waste products from cells. The classifier for blood-selective gene prediction showed overall accuracy at 93.29% with MCC = 0.5109 and ROC AUC = 0.9170. The kidneys play key roles in urinary system. The organs filter waste products from the blood and excrete them in urine. The classifier for kidney-selective gene prediction achieved predictive performance with overall accuracy at 93.62% with MCC = 0.4648 and ROC AUC = 0.9300. The results suggest that our approach can be used to identify the genes that have preferential expression in different types of tissues.
A machine learning approach has been developed in this study for identifying the human tissue-specific gene targets. Random Forests (RFs) and Support Vector Machines (SVMs) were trained separately with the microarray gene expression data to construct classifiers for prediction. It was shown that the RF classifiers outperform SVM models for tissue-specific gene prediction. 1,408 brain-specific gene targets and 493 liver-specific gene targets were identified using RF classifiers. The predicted targets show clear expression patterns in corresponding tissue samples, and they have functions and regulatory elements consistent with the tissues in GO enrichment analysis and transcription factor prediction. The analysis of high-scoring candidate genes for brain and liver specific expression suggests that our approach can select some interesting targets for further experimental studies. Our approach could also provide useful information for tissue-selective gene prediction. The approach can be used to develop new drug targets for biomedical research and expand our knowledge of tissue-specific expression.
This work is supported by the CSREES/USDA, under project number SC-1700355.
This article has been published as part of BMC Medical Genomics Volume 6 Supplement 1, 2013: Proceedings of the 2011 International Conference on Bioinformatics and Computational Biology (BIOCOMP'11). The full contents of the supplement are available online at http://www.biomedcentral.com/bmcmedgenomics/supplements/6/S1. Publication of this supplement has been supported by the International Society of Intelligent Biological Medicine.
- Lage K, Hansen NT, Karlberg EO, Eklund AC, Roque FS, Donahoe PK, Szallasi Z, Jensen TS, Brunak S: A large-scale analysis of tissue-specific pathology and gene expression of human disease genes and complexes. Proc Natl Acad Sci USA. 2008, 105 (52): 20870-20875. 10.1073/pnas.0810772105.PubMed CentralView ArticlePubMedGoogle Scholar
- Saltiel AR, Kahn CR: Insulin signalling and the regulation of glucose and lipid metabolism. Nature. 2001, 414 (6865): 799-806. 10.1038/414799a.View ArticlePubMedGoogle Scholar
- Liotta LA, Kohn EC: The microenvironment of the tumour-host interface. Nature. 2001, 411 (6835): 375-379. 10.1038/35077241.View ArticlePubMedGoogle Scholar
- Goh KI, Cusick ME, Valle D, Childs B, Vidal M, Barabasi AL: The human disease network. Proc Natl Acad Sci USA. 2007, 104 (21): 8685-8690. 10.1073/pnas.0701361104.PubMed CentralView ArticlePubMedGoogle Scholar
- Kadota K, Nishimura S, Bono H, Nakamura S, Hayashizaki Y, Okazaki Y, Takahashi K: Detection of genes with tissue-specific expression patterns using Akaike's information criterion procedure. Physiol Genomics. 2003, 12 (3): 251-259.View ArticlePubMedGoogle Scholar
- Kadota K, Ye J, Nakai Y, Terada T, Shimizu K: ROKU: a novel method for identification of tissue-specific genes. BMC Bioinformatics. 2006, 7: 294-10.1186/1471-2105-7-294.PubMed CentralView ArticlePubMedGoogle Scholar
- Liang S, Li Y, Be X, Howes S, Liu W: Detecting and profiling tissue-selective genes. Physiol Genomics. 2006, 26 (2): 158-162. 10.1152/physiolgenomics.00313.2005.View ArticlePubMedGoogle Scholar
- Chikina MD, Huttenhower C, Murphy CT, Troyanskaya OG: Global prediction of tissue-specific gene expression and context-dependent gene networks in Caenorhabditis elegans. PLoS Comput Biol. 2009, 5 (6): e1000417-10.1371/journal.pcbi.1000417.PubMed CentralView ArticlePubMedGoogle Scholar
- Wang L, Srivastava AK, Schwartz CE: Microarray data integration for genome-wide analysis of human tissue-selective gene expression. BMC Genomics. 2010, 11 (Suppl 2): S15-10.1186/1471-2164-11-S2-S15.PubMed CentralView ArticlePubMedGoogle Scholar
- Li C, Hung Wong W: Model-based analysis of oligonucleotide arrays: model validation, design issues and standard error application. Genome Biol. 2001, 2 (8): RESEARCH0032-PubMed CentralPubMedGoogle Scholar
- Li C, Wong WH: Model-based analysis of oligonucleotide arrays: expression index computation and outlier detection. Proc Natl Acad Sci USA. 2001, 98 (1): 31-36. 10.1073/pnas.98.1.31.PubMed CentralView ArticlePubMedGoogle Scholar
- Liaw A, Wiener M: Classification and Regression by randomForest. R News. 2002, 2 (3): 18-22.Google Scholar
- Noble WS: What is a support vector machine?. Nat Biotechnol. 2006, 24 (12): 1565-1567. 10.1038/nbt1206-1565.View ArticlePubMedGoogle Scholar
- Joachims T: Learning to classify text using support vector machines. 2002, Boston: Kluwer Academic PublishersView ArticleGoogle Scholar
- Swets JA: Measuring the accuracy of diagnostic systems. Science. 1988, 240 (4857): 1285-1293. 10.1126/science.3287615.View ArticlePubMedGoogle Scholar
- Bradley A: The use of the area under the ROC curve in the evaluation of machine learning algorithms. Pattern Recognition. 1997, 30: 1145-1159. 10.1016/S0031-3203(96)00142-2.View ArticleGoogle Scholar
- Dennis G, Sherman BT, Hosack DA, Yang J, Gao W, Lane HC, Lempicki RA: DAVID: Database for Annotation, Visualization, and Integrated Discovery. Genome Biol. 2003, 4 (5): P3-10.1186/gb-2003-4-5-p3.View ArticlePubMedGoogle Scholar
- Carlson JM, Chakravarty A, DeZiel CE, Gross RH: SCOPE: a web server for practical de novo motif discovery. Nucleic Acids Res. 2007, 35 (Web Server): W259-264. 10.1093/nar/gkm310.PubMed CentralView ArticlePubMedGoogle Scholar
- Gotea V, Ovcharenko I: DiRE: identifying distant regulatory elements of co-expressed genes. Nucleic Acids Res. 2008, 36 (Web Server): W133-139. 10.1093/nar/gkn300.PubMed CentralView ArticlePubMedGoogle Scholar
- Saeed AI, Bhagabati NK, Braisted JC, Liang W, Sharov V, Howe EA, Li J, Thiagarajan M, White JA, Quackenbush J: TM4 microarray software suite. Methods Enzymol. 2006, 411: 134-193.View ArticlePubMedGoogle Scholar
- Fujii H, Hamada H: A CNS-specific POU transcription factor, Brn-2, is required for establishing mammalian neural cell lineages. Neuron. 1993, 11 (6): 1197-1206. 10.1016/0896-6273(93)90231-F.View ArticlePubMedGoogle Scholar
- Yamanaka T, Tosaki A, Miyazaki H, Kurosawa M, Furukawa Y, Yamada M, Nukina N: Mutant huntingtin fragment selectively suppresses Brn-2 POU domain transcription factor to mediate hypothalamic cell dysfunction. Hum Mol Genet. 2010, 19 (11): 2099-2112. 10.1093/hmg/ddq087.PubMed CentralView ArticlePubMedGoogle Scholar
- Hayashi A, Kasahara T, Iwamoto K, Ishiwata M, Kametani M, Kakiuchi C, Furuichi T, Kato T: The role of brain-derived neurotrophic factor (BDNF)-induced XBP1 splicing during brain development. J Biol Chem. 2007, 282 (47): 34525-34534. 10.1074/jbc.M704300200.View ArticlePubMedGoogle Scholar
- Ma K, Zheng S, Zuo Z: The transcription factor regulatory factor X1 increases the expression of neuronal glutamate transporter type 3. J Biol Chem. 2006, 281 (30): 21250-21255. 10.1074/jbc.M600521200.View ArticlePubMedGoogle Scholar
- Tedeschi A, Di Giovanni S: The non-apoptotic role of p53 in neuronal biology: enlightening the dark side of the moon. EMBO Rep. 2009, 10 (6): 576-583. 10.1038/embor.2009.89.PubMed CentralView ArticlePubMedGoogle Scholar
- Anant S, Henderson JO, Mukhopadhyay D, Navaratnam N, Kennedy S, Min J, Davidson NO: Novel role for RNA-binding protein CUGBP2 in mammalian RNA editing. CUGBP2 modulates C to U editing of apolipoprotein B mRNA by interacting with apobec-1 and ACF, the apobec-1 complementation factor. J Biol Chem. 2001, 276 (50): 47338-47351. 10.1074/jbc.M104911200.View ArticlePubMedGoogle Scholar
- Mukhopadhyay D, Houchen CW, Kennedy S, Dieckgraefe BK, Anant S: Coupled mRNA stabilization and translational silencing of cyclooxygenase-2 by a novel RNA binding protein, CUGBP2. Mol Cell. 2003, 11 (1): 113-126. 10.1016/S1097-2765(03)00012-1.View ArticlePubMedGoogle Scholar
- Yang Y, Mahaffey CL, Berube N, Maddatu TP, Cox GA, Frankel WN: Complex seizure disorder caused by Brunol4 deficiency in mice. PLoS Genet. 2007, 3 (7): e124-10.1371/journal.pgen.0030124.PubMed CentralView ArticlePubMedGoogle Scholar
- Ghersi E, Noviello C, D'Adamio L: Amyloid-beta protein precursor (AbetaPP) intracellular domain-associated protein-1 proteins bind to AbetaPP and modulate its processing in an isoform-specific manner. J Biol Chem. 2004, 279 (47): 49105-49112. 10.1074/jbc.M405329200.View ArticlePubMedGoogle Scholar
- Tanji K, Kamitani T, Mori F, Kakita A, Takahashi H, Wakabayashi K: TRIM9, a novel brain-specific E3 ubiquitin ligase, is repressed in the brain of Parkinson's disease and dementia with Lewy bodies. Neurobiol Dis. 2010, 38 (2): 210-218. 10.1016/j.nbd.2010.01.007.PubMed CentralView ArticlePubMedGoogle Scholar
- Rauch U, Karthikeyan L, Maurel P, Margolis RU, Margolis RK: Cloning and primary structure of neurocan, a developmentally regulated, aggregating chondroitin sulfate proteoglycan of brain. J Biol Chem. 1992, 267 (27): 19536-19547.PubMedGoogle Scholar
- Somia NV, Schmitt MJ, Vetter DE, Van Antwerp D, Heinemann SF, Verma IM: LFG: an anti-apoptotic gene that provides protection from Fas-mediated cell death. Proc Natl Acad Sci USA. 1999, 96 (22): 12667-12672. 10.1073/pnas.96.22.12667.PubMed CentralView ArticlePubMedGoogle Scholar
- Reed JE, Dunn JR, du Plessis DG, Shaw EJ, Reeves P, Gee AL, Warnke PC, Sellar GC, Moss DJ, Walker C: Expression of cellular adhesion molecule 'OPCML' is down-regulated in gliomas and other brain tumours. Neuropathol Appl Neurobiol. 2007, 33 (1): 77-85.View ArticlePubMedGoogle Scholar
- Tipsmark CK: Identification of FXYD protein genes in a teleost: tissue-specific expression and response to salinity change. Am J Physiol Regul Integr Comp Physiol. 2008, 294 (4): R1367-1378. 10.1152/ajpregu.00454.2007.View ArticlePubMedGoogle Scholar
- Kest B, Smith SB, Schorscher-Petcu A, Austin JS, Ritchie J, Klein G, Rossi GC, Fortin A, Mogil JS: Gnao1 (G alphaO protein) is a likely genetic contributor to variation in physical dependence on opioids in mice. Neuroscience. 2009, 162 (4): 1255-1264. 10.1016/j.neuroscience.2009.05.027.View ArticlePubMedGoogle Scholar
- Li L, Chin LS, Greengard P, Copeland NG, Gilbert DJ, Jenkins NA: Localization of the synapsin II (SYN2) gene to human chromosome 3 and mouse chromosome 6. Genomics. 1995, 28 (2): 365-366. 10.1006/geno.1995.1162.View ArticlePubMedGoogle Scholar
- Portela-Gomes GM, Lukinius A, Grimelius L: Synaptic vesicle protein 2, a new neuroendocrine cell marker. Am J Pathol. 2000, 157 (4): 1299-1309. 10.1016/S0002-9440(10)64645-7.PubMed CentralView ArticlePubMedGoogle Scholar
- Adams JH, Wigg KG, King N, Burcescu I, Vetro A, Kiss E, Baji I, George CJ, Kennedy JL, Kovacs M, et al: Association study of neurotrophic tyrosine kinase receptor type 2 (NTRK2) and childhood-onset mood disorders. Am J Med Genet B Neuropsychiatr Genet. 2005, 132B (1): 90-95. 10.1002/ajmg.b.30084.View ArticlePubMedGoogle Scholar
- Kamimura K, Fujise M, Villa F, Izumi S, Habuchi H, Kimata K, Nakato H: Drosophila heparan sulfate 6-O-sulfotransferase (dHS6ST) gene. Structure, expression, and function in the formation of the tracheal system. J Biol Chem. 2001, 276 (20): 17014-17021. 10.1074/jbc.M011354200.View ArticlePubMedGoogle Scholar
- Sugawara T, Tsurubuchi Y, Agarwala KL, Ito M, Fukuma G, Mazaki-Miyazaki E, Nagafuji H, Noda M, Imoto K, Wada K, et al: A missense mutation of the Na+ channel alpha II subunit gene Na(v)1.2 in a patient with febrile and afebrile seizures causes channel dysfunction. Proc Natl Acad Sci USA. 2001, 98 (11): 6384-6389. 10.1073/pnas.111065098.PubMed CentralView ArticlePubMedGoogle Scholar
- Szabo G, Romics L, Frendl G: Liver in sepsis and systemic inflammatory response syndrome. Clin Liver Dis. 2002, 6 (4): 1045-1066. 10.1016/S1089-3261(02)00058-2. xView ArticlePubMedGoogle Scholar
- Schrem H, Klempnauer J, Borlak J: Liver-enriched transcription factors in liver function and development. Part I: the hepatocyte nuclear factor network and liver-specific gene expression. Pharmacol Rev. 2002, 54 (1): 129-158. 10.1124/pr.54.1.129.View ArticlePubMedGoogle Scholar
- Kalafatis M, Egan JO, van't Veer C, Cawthern KM, Mann KG: The regulation of clotting factors. Crit Rev Eukaryot Gene Expr. 1997, 7 (3): 241-280. 10.1615/CritRevEukarGeneExpr.v7.i3.40.View ArticlePubMedGoogle Scholar
- Zhang T, Yao S, Wang P, Yin C, Xiao C, Qian M, Liu D, Zheng L, Meng W, Zhu H, et al: Apoa-Ii directs morphogenetic movements of zebrafish embryo by preventing chromosome fusion during nuclear division in yolk syncytial layer. J Biol Chem. 2011Google Scholar
- Charbonneau A, The VL: Genomic organization of a human 5beta-reductase and its pseudogene and substrate selectivity of the expressed enzyme. Biochim Biophys Acta. 2001, 1517 (2): 228-235. 10.1016/S0167-4781(00)00278-5.View ArticlePubMedGoogle Scholar
- Boomgaarden I, Vock C, Klapper M, Doring F: Comparative analyses of disease risk genes belonging to the acyl-CoA synthetase medium-chain (ACSM) family in human liver and cell lines. Biochem Genet. 2009, 47 (9-10): 739-748. 10.1007/s10528-009-9273-z.View ArticlePubMedGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.