Multiple imputation for missing continuous and dichotomous data: The effects of patterns of missingness and variable correlations on type I error with small samples

The purpose of this study is to investigate the effects of predictor variable correlations and patterns of missingness with dichotomous and/or continuous data in small samples when missing data is multiply imputed. Missing data of predictor variables is multiply imputed under three different multivariate models: the multivariate normal model for continuous data, the multinomial model for dichotomous data and the general location model for mixed dichotomous and continuous data. Subsequent to the multiple imputation process, Type I error rates of the regression coefficients obtained with logistic regression analysis are estimated under various conditions of correlation structure, sample size, type of data and patterns of missing data. The distributional properties of average mean, variance and correlations among the predictor variables are assessed after the multiple imputation process. ^ For continuous predictor data under the multivariate normal model, Type I error rates are generally within the nominal values with samples of size n = 100. Smaller samples of size n = 50 resulted in more conservative estimates (i.e., lower than the nominal value). Correlation and variance estimates of the original data are retained after multiple imputation with less than 50% missing continuous predictor data. For dichotomous predictor data under the multinomial model, Type I error rates are generally conservative, which in part is due to the sparseness of the data. The correlation structure for the predictor variables is not well retained on multiply-imputed data from small samples with more than 50% missing data with this model. For mixed continuous and dichotomous predictor data, the results are similar to those found under the multivariate normal model for continuous data and under the multinomial model for dichotomous data. With all data types, a fully-observed variable included with variables subject to missingness in the multiple imputation process and subsequent statistical analysis provided liberal (larger than nominal values) Type I error rates under a specific pattern of missing data. It is suggested that future studies focus on the effects of multiple imputation in multivariate settings with more realistic data characteristics and a variety of multivariate analyses, assessing both Type I error and power. ^

Comparison of multiple comparison methods for identifying differential gene expression in simulated and real papillary thyroid cancer microarray data

The difficulty of detecting differential gene expression in microarray data has existed for many years. Several correction procedures try to avoid the family-wise error rate in multiple comparison process, including the Bonferroni and Sidak single-step p-value adjustments, Holm's step-down correction method, and Benjamini and Hochberg's false discovery rate (FDR) correction procedure. Each multiple comparison technique has its advantages and weaknesses. We studied each multiple comparison method through numerical studies (simulations) and applied the methods to the real exploratory DNA microarray data, which detect of molecular signatures in papillary thyroid cancer (PTC) patients. According to our results of simulation studies, Benjamini and Hochberg step-up FDR controlling procedure is the best process among these multiple comparison methods and we discovered 1277 potential biomarkers among 54675 probe sets after applying the Benjamini and Hochberg's method to PTC microarray data.^

THE MULTIPLE LOCATION TIME WEIGHTED INDEX: USING PATIENT ACTIVITY SPACES TO CALCULATE PRIMARY CARE SERVICE AREAS

Geographic health planning analyses, such as service area calculations, are hampered by a lack of patient-specific geographic data. Using the limited patient address information in patient management systems, planners analyze patient origin based on home address. But activity space research done sparingly in public health and extensively in non-health related arenas uses multiple addresses per person when analyzing accessibility. Also, health care access research has shown that there are many non-geographic factors that influence choice of provider. Most planning methods, however, overlook non-geographic factors influencing choice of provider, and the limited data mean the analyses can only be related to home address. This research attempted to determine to what extent geography plays a part in patient choice of provider and to determine if activity space data can be used to calculate service areas for primary care providers. During Spring 2008, a convenience sample of 384 patients of a locally-funded Community Health Center in Houston, Texas, completed a survey that asked about what factors are important when he or she selects a health care provider. A subset of this group (336) also completed an activity space log that captured location and time data on the places where the patient regularly goes. Survey results indicate that for this patient population, geography plays a role in their choice of health care provider, but it is not the most important reason for choosing a provider. Other factors for choosing a health care provider such as the provider offering “free or low cost visits”, meeting “all of the patient’s health care needs”, and seeing “the patient quickly” were all ranked higher than geographic reasons. Analysis of the patient activity locations shows that activity spaces can be used to create service areas for a single primary care provider. Weighted activity-space-based service areas have the potential to include more patients in the service area since more than one location per patient is used. Further analysis of the logs shows that a reduced set of locations by time and type could be used for this methodology, facilitating ongoing data collection for activity-space-based planning efforts.

Multiple independent genetic factors at NOS1AP modulate the QT interval in a multi-ethnic population.

Extremes of electrocardiographic QT interval are associated with increased risk for sudden cardiac death (SCD); thus, identification and characterization of genetic variants that modulate QT interval may elucidate the underlying etiology of SCD. Previous studies have revealed an association between a common genetic variant in NOS1AP and QT interval in populations of European ancestry, but this finding has not been extended to other ethnic populations. We sought to characterize the effects of NOS1AP genetic variants on QT interval in the multi-ethnic population-based Dallas Heart Study (DHS, n = 3,072). The SNP most strongly associated with QT interval in previous samples of European ancestry, rs16847548, was the most strongly associated in White (P = 0.005) and Black (P = 3.6 x 10(-5)) participants, with the same direction of effect in Hispanics (P = 0.17), and further showed a significant SNP x sex-interaction (P = 0.03). A second SNP, rs16856785, uncorrelated with rs16847548, was also associated with QT interval in Blacks (P = 0.01), with qualitatively similar results in Whites and Hispanics. In a previously genotyped cohort of 14,107 White individuals drawn from the combined Atherosclerotic Risk in Communities (ARIC) and Cardiovascular Health Study (CHS) cohorts, we validated both the second locus at rs16856785 (P = 7.63 x 10(-8)), as well as the sex-interaction with rs16847548 (P = 8.68 x 10(-6)). These data extend the association of genetic variants in NOS1AP with QT interval to a Black population, with similar trends, though not statistically significant at P<0.05, in Hispanics. In addition, we identify a strong sex-interaction and the presence of a second independent site within NOS1AP associated with the QT interval. These results highlight the consistent and complex role of NOS1AP genetic variants in modulating QT interval.

MULTIPLE POSTTRANSCRIPTIONAL REGULATORY FEATURES CONTROL EXPRESSION OF ETHANOLAMINE UTILIZATION GENES IN ENTEROCOCCUS FAECALIS

Enterococcus faecalis is a Gram-positive bacterium that lives as a commensal organism in the mammalian gastrointestinal tract, but can behave as an opportunistic pathogen. Our lab discovered that mutation of the eutK gene attenuates virulence of E. faecalis in the C. elegans model host. eutK is part of the ethanolamine metabolic pathway which was previously unknown in E. faecalis. I discovered the presence of two unique posttranscriptional regulatory features that control expression of eut locus genes. The first feature I found is an AdoCBL riboswitch, a cis-acting RNA regulatory element that acts as a positive regulator of gene expression. The second feature I discovered is a unique two-component system, EutVW. The EutV response regulator contains an ANTAR family domain, which binds RNA to trigger transcriptional antitermination. I determined that induction of expression of several genes in the eut locus is dependent on ethanolamine, AdoCBL and the two-component system. AdoCBL and ethanolamine are both required for induction of eut locus gene expression. Additionally, I discovered eutG is regulated by a unique mechanism of antitermination. Both the AdoCBL riboswitch and EutV response regulator control the expression of the downstream gene eutG. EutV potentially acts through a novel antitermination mechanism in which a dimer of EutV binds to a pair of mRNA stem loops forming an antitermination complex. My data show a unique mechanism by which two environmental signals are integrated by two different posttranscriptional regulators to regulate a single locus.

BRAIN ACTIVATION AND CONNECTIVITY IN NON-DISABLED MULTIPLE SCLEROSIS PATIENTS

Multiple sclerosis (MS) is the most common demyelinating disease affecting the central nervous system. There is no cure for MS and current therapies have limited efficacy. While the majority of individuals with MS develop significant clinical disability, a subset experiences a disease course with minimal impairment even in the presence of significant apparent tissue damage on magnetic resonance imaging (MRI). The current studies combined functional MRI and diffusion tensor imaging (DTI) to elucidate brain mechanisms associated with lack of clinical disability in patients with MS. Recent evidence has implicated cortical reorganization as a mechanism to limit the clinical manifestation of the disease. Functional MRI was used to test the hypothesis that non-disabled MS patients (Expanded Disability Status Scale ≤ 1.5) show increased recruitment of cognitive control regions (dorsolateral prefrontal and anterior cingulate cortex) while performing sensory, motor and cognitive tasks. Compared to matched healthy controls, patients increased activation of cognitive control brain regions when performing non-dominant hand movements and the 2-back working memory task. Using dynamic causal modeling, we tested whether increased cognitive control recruitment is associated with alterations in connectivity in the working memory functional network. Patients exhibited similar network connectivity to that of control subjects when performing working memory tasks. We subsequently investigated the integrity of major white matter tracts to assess structural connectivity and its relation to activation and functional integration of the cognitive control system. Patients showed substantial alterations in callosal, inferior and posterior white matter tracts and less pronounced involvement of the corticospinal tracts and superior longitudinal fasciculi (SLF). Decreased structural integrity within the right SLF in patients was associated with decreased performance, and decreased activation and connectivity of the cognitive control system when performing working memory tasks. These studies suggest that patient with MS without clinical disability increase cognitive control system recruitment across functional domains and rely on preserved functional and structural connectivity of brain regions associated with this network. Moreover, the current studies show the usefulness of combining brain activation data from functional MRI and structural connectivity data from DTI to improve our understanding of brain adaptation mechanisms to neurological disease.

Multiple site phosphorylation of tyrosine hydroxylase: A potential role for protein kinase C in the regulation of catecholamine synthesis

Tyrosine hydroxylase (TH), the initial and rate limiting enzyme in the catecholaminergic biosynthetic pathway, is phosphorylated on multiple serine residues by multiple protein kinases. Although it has been demonstrated that many protein kinases are capable of phosphorylating and activating TH in vitro, it is less clear which protein kinases participate in the physiological regulation of catecholamine synthesis in situ. These studies were designed to determine if protein kinase C (PK-C) plays such a regulatory role.^ Stimulation of intact bovine adrenal chromaffin cells with phorbol esters results in stimulation of catecholamine synthesis, tyrosine hydroxylase phosphorylation and activation. These responses are both time and concentration dependent, and are specific for those phorbol ester analogues which activate PK-C. RP-HPLC analysis of TH tryptic phosphopeptides indicate that PK-C phosphorylates TH on three putative sites. One of these (pepetide 6) is the same as that phosphorylated by both cAMP-dependent protein kinase (PK-A) and calcium/calmodulin-dependent protein kinase (CaM-K). However, two of these sites (peptides 4 and 7) are unique, and, to date, have not been shown to be phosphorylated by any other protein kinase. These peptides correspond to those which are phosphorylated with a slow time course in response to stimulation of chromaffin cells with the natural agonist acetylcholine. The activation of TH produced by PK-C is most closely correlated with the phosphorylation of peptide 6. But, as evident from pH profiles of tyrosine hydroxylase activity, phosphorylation of peptides 4 and 7 affect the expression of the activation produced by phosphorylation of peptide 6.^ These data support a role for PK-C in the control of TH activity, and suggest a two stage model for the physiological regulation of catecholamine synthesis by phosphorylation in response to cholinergic stimulation. An initial fast response, which appears to be mediated by CaM-K, and a slower, sustained response which appears to be mediated by PK-C. In addition, the multiple site phosphorylation of TH provides a mechanism whereby the regulation of catecholamine synthesis appears to be under the control of multiple protein kinases, and allows for the convergence of multiple, diverse physiological and biochemical signals. ^

Multiple GABA receptors in rabbit retina

The task of encoding and processing complex sensory input requires many types of transsynaptic signals. This requirement is served in part by an extensive group of neurotransmitter substances which may include thirty or more different compounds. At the next level of information processing, the existence of multiple receptors for a given neurotransmitter appears to be a widely used mechanism to generate multiple responses to a given first messenger (Snyder and Goodman, 1980). Despite the wealth of published data on GABA receptors, the existence of more than one GABA receptor was in doubt until the mid 1980's. Presently there is still disagreement on the number of types of GABA receptors, estimates for which range from two to four (DeFeudis, 1983; Johnston, 1985). Part of the problem in evaluating data concerning multiple receptor types is the lack of information on the number of gene products and their subsequent supramolecular organization in different neurons. In order to evaluate the question concerning the diversity of GABA receptors in the nervous system, we must rely on indirect information derived from a wide variety of experimental techniques. These include pharmacological binding studies to membrane fractions, electrophysiological studies, localization studies, purification studies, and functional assays. Almost all parts of the central and peripheral nervous system use GABA as a neurotransmitter, and these experimental techniques have therefore been applied to many different parts of the nervous system for the analysis of GABA receptor characteristics. We are left with a large amount of data from a wide variety of techniques derived from many parts of the nervous system. When this project was initiated in 1983, there were only a handful of pharmacological tools to assess the question of multiple GABA receptors. The approach adopted was to focus on a single model system, using a variety of experimental techniques, in order to evaluate the existence of multiple forms of GABA receptors. Using the in vitro rabbit retina, a combination of pharmacological binding studies, functional release studies and partial purification studies were undertaken to examine the GABA receptor composition of this tissue. Three types of GABA receptors were observed: Al receptors coupled to benzodiazepine and barbiturate modulation, and A2 or uncoupled GABA-A receptors, and GABA-B receptors. These results are evaluated and discussed in light of recent findings by others concerning the number and subtypes of GABA receptors in the nervous system. ^

CYTOGENETIC AND DNA CONTENT ANALYSIS IN MULTIPLE MYELOMA

In this dissertation, the cytogenetic characteristics of bone marrow cells from 41 multiple myeloma patients were investigated. These cytogenetic data were correlated with the total DNA content as measured by flow cytometry. Both the cytogenetic information and DNA content were then correlated with clinical data to determine if diagnosis and prognosis of multiple myeloma could be improved.^ One hundred percent of the patients demonstrated abnormal chromosome numbers per metaphase. The average chromosome number per metaphase ranged from 42 to 49.9, with a mean of 44.99. The percent hypodiploidy ranged from 0-100% and the percent hyperdiploidy from 0-53%. Detailed cytogenetic analyses were very difficult to perform because of the paucity of mitotic figures and the poor chromosome morphology. Thus, detailed chromosome banding analysis on these patients was impossible.^ Thirty seven percent of the patients had normal total DNA content, whereas 63% had abnormal amounts of DNA (one patient with less than normal amounts and 25 patients with greater than normal amounts of DNA).^ Several clinical parameters were used in the statistical analyses: tumor burden, patient status at biopsy, patient response status, past therapy, type of treatment and percent plasma cells. Only among these clinical parameters were any statistically significant correlations found: pretreatment tumor burden versus patient response, patient biopsy status versus patient response and past therapy versus patient response.^ No correlations were found between percent hypodiploid, diploid, hyperdiploid or DNA content, and the patient response status, nor were any found between those patients with: (a) normal plasma cells, low pretreatment tumor mass burden and more than 50% of the analyzed metaphases with 46 chromosomes; (b) normal amounts of DNA, low pretreatment tumor mass burden and more than 50% of the metaphases with 46 chromosomes; (c) normal amounts of DNA and normal quantities of plasma cells; (d) abnormal amounts of DNA, abnormal amounts of plasma cells, high pretreatment tumor mass burden and less than 50% of the metaphases with 46 chromosomes.^ Technical drawbacks of both cytogenetic and DNA content analysis in these multiple myeloma patients are discussed along with the lack of correlations between DNA content and chromosome number. Refined chromosome banding analysis awaits technical improvements before we can understand which chromosome material (if any) makes up the "extra" amounts of DNA in these patients. None of the correlations tested can be used as diagnostic or prognostic aids for multiple myeloma. ^

14-3-3zeta overexpression in multiple human cancers plays a critical role in cancer progression

14-3-3 is a family of highly conserved and ubiquitously expressed proteins in eukaryotic organisms. 14-3-3 isoforms bind in a phospho-serine/threonine-dependent manner to a host of proteins involved in essential cellular processes including cell cycle, signal transduction and apoptosis. We fortuitously discovered 14-3-3 zeta overexpression in many human primary cancers, such as breast, lung, and sarcoma, and in a majority of cancer cell lines. To determine 14-3-3 zeta involvement in breast cancer progression, we used immunohistochemical analysis to examine 14-3-3 zeta expression in human primary invasive breast carcinomas. High 14-3-3 zeta expression was significantly correlated with poor prognosis of breast cancer patients. Increased expression of 14-3-3 zeta was also significantly correlated with elevated PKB/Akt activation in patient samples. Thus, 14-3-3 zeta is a marker of poor prognosis in breast cancers. Furthermore, up-regulation of 14-3-3 zeta enhanced malignant transformation of cancer cells in vitro. ^ To determine the biological significance of 14-3-3 zeta in human cancers, small interfering RNAs (siRNA) were used to specifically block 14-3-3 zeta expression in cancer cells. 14-3-3 zeta siRNA inhibited cellular proliferation by inducing a G1 arrest associated with up-regulation of p27 KIP1 and p21CIP1 cyclin dependent kinase inhibitors. Reduced 14-3-3 zeta inhibited PKB/Akt activation while stimulating the p38 signaling pathway. Silencing 14-3-3 zeta expression also increased stress-induced apoptosis by caspase activation. Notably, 14-3-3 zeta siRNA inhibited transformation related properties of breast cancer cells in vitro and inhibited tumor progression of breast cancer cells in vivo. 14-3-3 zeta may be a key regulatory factor controlling multiple signaling pathways leading to tumor progression. ^ The data indicate 14-3-3 zeta is a major regulator of cell growth and apoptosis and may play a critical role in the development of multiple cancer types. Hence, blocking 14-3-3 zeta may be a promising therapeutic approach for numerous cancers. ^

Obesity and risk of multiple myeloma: A case-control study

Introduction. Several studies have reported a positive association of body mass index (BMI) with multiple myeloma; however, the period of adulthood where BMI is most important remains unclear. In addition, it is well known that body fat is associated with both sex-steroid hormone storage and with increasing insulin levels; therefore, it was hypothesized that the association between obesity and multiple myeloma may be attributed to increased aromatization of androgen in adipose tissue. Objective. The overall objective of this case-control study was to determine whether multiple myeloma cases had higher BMI and greater adult weight gain relative to healthy controls. In addition, we tested the hypothesis that hormone replacement therapy use among women will further increase the association between BMI and risk of multiple myeloma. This study used data from a pilot case-control study at M.D. Anderson Cancer Center (MDACC), entitled Etiology of multiple myeloma, directed by Dr. Sara Strom and Dr. Sergio Giralt. Methods. The pilot study recruited a total of 122 cases of histopathologically confirmed multiple myeloma from MDACC. Controls (n=183) were selected from a database of random digit dialing controls accrued in the Department of Epidemiology at MDACC and were frequency matched to the cases on age (±5 years), gender, and race/ethnicity. Demographic and risk factor information were obtained from all participants who completed a self-administered questionnaire. Items included in the questionnaire include demographic information, height and weight at age 25, 40 and current/diagnosis, medical history, family history of cancer, smoking and alcohol use. Statistical analysis. Initial descriptive analysis included Student's t-test and Pearson's chi-squared tests. Odds ratios and 95% confidence intervals were calculated to quantify the association between the variables of interest and multiple myeloma. A multivariable model will be developed using unconditional logistic regression. Results. MM cases were 1.79 times (95% CI=0.99-3.32) more likely to have been overweight or obese (BMI > 25 kg/m2) at age 25 relative to healthy controls after controlling for age, gender, race/ethnicty, education and family history of cancer. Being overweight or obese at age 40 was not significantly associated with mutliple myeloma risk (OR=1.42, 95% CI=0.86-2.34) nor was being overweight or obses at diagnosis (OR=1.43, 95% CI=0.78, 2.63). We observed a statistically significant 2-fold increased odds of multiple myeloma in individuals who gained more than 4.7 kg during between 25 and 40 years (OR=1.97, 95% CI=1.15-3.39). When assessing HRT as a modifier of the BMI and multiple myeloma association among women (N=123), no association between obesity and MM status was observed among women who have never used HRT (OR=0.60, 95% CI=0.23-1.61; n=73). Yet among women who have ever used HRT (n=50), being overweight or obese was associated with an increase in MM risk (OR=2. 93, 95% CI=0.81-10.6) after adjusting for age; however, the association was not statistically significant. Significance. This study provides further evidence that increased BMI increases the risk of multiple myeloma. Furthermore, among women, HRT use may modify risk of disease. ^

The multiple location time weighted index: Using patient activity spaces to calculate primary care service areas

Geographic health planning analyses, such as service area calculations, are hampered by a lack of patient-specific geographic data. Using the limited patient address information in patient management systems, planners analyze patient origin based on home address. But activity space research done sparingly in public health and extensively in non-health related arenas uses multiple addresses per person when analyzing accessibility. Also, health care access research has shown that there are many non-geographic factors that influence choice of provider. Most planning methods, however, overlook non-geographic factors influencing choice of provider, and the limited data mean the analyses can only be related to home address. This research attempted to determine to what extent geography plays a part in patient choice of provider and to determine if activity space data can be used to calculate service areas for primary care providers. ^ During Spring 2008, a convenience sample of 384 patients of a locally-funded Community Health Center in Houston, Texas, completed a survey that asked about what factors are important when he or she selects a health care provider. A subset of this group (336) also completed an activity space log that captured location and time data on the places where the patient regularly goes. ^ Survey results indicate that for this patient population, geography plays a role in their choice of health care provider, but it is not the most important reason for choosing a provider. Other factors for choosing a health care provider such as the provider offering "free or low cost visits", meeting "all of the patient's health care needs", and seeing "the patient quickly" were all ranked higher than geographic reasons. ^ Analysis of the patient activity locations shows that activity spaces can be used to create service areas for a single primary care provider. Weighted activity-space-based service areas have the potential to include more patients in the service area since more than one location per patient is used. Further analysis of the logs shows that a reduced set of locations by time and type could be used for this methodology, facilitating ongoing data collection for activity-space-based planning efforts. ^

Functional data analysis approaches for genotype-phenotype association studies from next-generation sequencing

Next-generation DNA sequencing platforms can effectively detect the entire spectrum of genomic variation and is emerging to be a major tool for systematic exploration of the universe of variants and interactions in the entire genome. However, the data produced by next-generation sequencing technologies will suffer from three basic problems: sequence errors, assembly errors, and missing data. Current statistical methods for genetic analysis are well suited for detecting the association of common variants, but are less suitable to rare variants. This raises great challenge for sequence-based genetic studies of complex diseases.^ This research dissertation utilized genome continuum model as a general principle, and stochastic calculus and functional data analysis as tools for developing novel and powerful statistical methods for next generation of association studies of both qualitative and quantitative traits in the context of sequencing data, which finally lead to shifting the paradigm of association analysis from the current locus-by-locus analysis to collectively analyzing genome regions.^ In this project, the functional principal component (FPC) methods coupled with high-dimensional data reduction techniques will be used to develop novel and powerful methods for testing the associations of the entire spectrum of genetic variation within a segment of genome or a gene regardless of whether the variants are common or rare.^ The classical quantitative genetics suffer from high type I error rates and low power for rare variants. To overcome these limitations for resequencing data, this project used functional linear models with scalar response to develop statistics for identifying quantitative trait loci (QTLs) for both common and rare variants. To illustrate their applications, the functional linear models were applied to five quantitative traits in Framingham heart studies. ^ This project proposed a novel concept of gene-gene co-association in which a gene or a genomic region is taken as a unit of association analysis and used stochastic calculus to develop a unified framework for testing the association of multiple genes or genomic regions for both common and rare alleles. The proposed methods were applied to gene-gene co-association analysis of psoriasis in two independent GWAS datasets which led to discovery of networks significantly associated with psoriasis.^

Statistical Methods for Differential Expressions of Genes Detected in Multiple-Condition Experiment of Microarray

Most studies of differential gene-expressions have been conducted between two given conditions. The two-condition experimental (TCE) approach is simple in that all genes detected display a common differential expression pattern responsive to a common two-condition difference. Therefore, the genes that are differentially expressed under the other conditions other than the given two conditions are undetectable with the TCE approach. In order to address the problem, we propose a new approach called multiple-condition experiment (MCE) without replication and develop corresponding statistical methods including inference of pairs of conditions for genes, new t-statistics, and a generalized multiple-testing method for any multiple-testing procedure via a control parameter C. We applied these statistical methods to analyze our real MCE data from breast cancer cell lines and found that 85 percent of gene-expression variations were caused by genotypic effects and genotype-ANAX1 overexpression interactions, which agrees well with our expected results. We also applied our methods to the adenoma dataset of Notterman et al. and identified 93 differentially expressed genes that could not be found in TCE. The MCE approach is a conceptual breakthrough in many aspects: (a) many conditions of interests can be conducted simultaneously; (b) study of association between differential expressions of genes and conditions becomes easy; (c) it can provide more precise information for molecular classification and diagnosis of tumors; (d) it can save lot of experimental resources and time for investigators.^