Differential genome-wide array-based methylation profiles in prognostic subsets of chronic lymphocytic leukemia.

Global hypomethylation and regional hypermethylation are well-known epigenetic features of cancer; however, in chronic lymphocytic leukemia (CLL), studies on genome-wide epigenetic modifications are limited. Here, we analyzed the global methylation profiles in CLL, by applying high-resolution methylation microarrays (27,578 CpG sites) to 23 CLL samples, belonging to the immunoglobulin heavy-chain variable (IGHV) mutated (favorable) and IGHV unmutated/IGHV3-21 (poor-prognostic) subsets. Overall, results demonstrated significant differences in methylation patterns between these subgroups. Specifically, in IGHV unmutated CLL, we identified methylation of 7 known or candidate tumor suppressor genes (eg, VHL, ABI3, and IGSF4) as well as 8 unmethylated genes involved in cell proliferation and tumor progression (eg, ADORA3 and PRF1 enhancing the nuclear factor-kappaB and mitogen-activated protein kinase pathways, respectively). In contrast, these latter genes were silenced by methylation in IGHV mutated patients. The array data were validated for selected genes using methylation-specific polymerase chain reaction, quantitative reverse transcriptase-polymerase chain reaction, and bisulfite sequencing. Finally, the significance of DNA methylation in regulating gene promoters was shown by reinducing 4 methylated tumor suppressor genes (eg, VHL and ABI3) in IGHV unmutated samples using the methyl-inhibitor 5-aza-2'-deoxycytidine. Taken together, our data for the first time reveal differences in global methylation profiles between prognostic subsets of CLL, which may unfold epigenetic silencing mechanisms involved in CLL pathogenesis.

Mantle cell lymphoma displays a homogenous methylation profile: a comparative analysis with chronic lymphocytic leukemia.

Mantle cell lymphoma (MCL) and chronic lymphocytic leukemia (CLL) are mature CD5(+) B-cell malignancies with different biological/clinical characteristics. We recently reported an association between different prognostic subgroups of CLL (i.e., IGHV mutated and unmutated) and genomic methylation pattern. However, the relationship between DNA methylation and prognostic markers, such as the proliferation gene expression signature, has not been investigated in MCL. We applied high-resolution methylation microarrays (27,578 CpG sites) to assess the global DNA methylation profiles in 20 MCL (10 each with high/low proliferation signature) and 30 CLL (15 poor-prognostic IGHV unmutated subset #1 and 15 good-prognostic IGHV mutated subset #4) samples. Notably, MCL and each CLL subset displayed distinct genomic methylation profiles. After unsupervised hierarchical clustering, 17/20 MCL cases formed a cluster separate from CLL, while CLL subsets #1 and #4 formed subclusters. Surprisingly, few differentially methylated genes (n = 6) were identified between high vs. low proliferation MCL. In contrast, distinct methylation profiles were demonstrated for MCL and CLL. Importantly, certain functional classes of genes were preferentially methylated in either disease. For instance, developmental genes, in particular homeobox transcription factor genes (e.g., HLXB9, HOXA13), were more highly methylated in MCL, whereas apoptosis-related genes were enriched among targets methylated in CLL (e.g., CYFIP2, NR4A1). Results were validated using pyrosequencing, RQ-PCR and reexpression of specific genes. In summary, the methylation profile of MCL was homogeneous and no correlation with the proliferation signature was observed. Compared to CLL, however, marked differences were discovered such as the preferential methylation of homeobox genes in MCL.

High-resolution genomic screening in mantle cell lymphoma--specific changes correlate with genomic complexity, the proliferation signature and survival.

Mantle cell lymphoma (MCL) is characterized by the t(11;14)(q13;q32) and numerous copy number aberrations (CNAs). Recently, gene expression profiling defined a proliferation gene expression signature in MCL where high scores predict shorter survival. We investigated 31 MCL cases using high-density single nucleotide polymorphism arrays and correlated CNA patterns with the proliferation signature and with clinical data. Many recurrent CNAs typical of MCL were detected, including losses at 1p (55%), 8p (29%), 9q (29%), 11q (55%), 13q (42%) and 17p (32%), and gains at 3q (39%), 8q (26%), 15q (23%) and 18q (23%). A novel deleted region at 20q (16%) contained only one candidate gene, ZFP64, a putative tumor suppressor. Unsupervised clustering identified subgroups with different patterns of CNAs, including a subset (19%) characterized by the presence of 11q loss in all cases and by the absence of 13q loss, and 3q and 7p gains. Losses at 1p, 8p, 13q and 17p were associated with increased genomic complexity. High proliferation signature scores correlated with increased number of large (>15 Mbp) CNAs (P = 0.03) as well as copy number gains at 7p (P = 0.02) and losses at 9q (P = 0.04). Furthermore, large/complex 13q losses were associated with improved survival (P < 0.05) as were losses/copy number neutral LOH at 19p13 (P = 0.01). In summary, this high-resolution genomic analysis identified novel aberrations and revealed that several CNAs correlated with genomic complexity, the proliferation status and survival.

Mechanisms underlying the sparing of masticatory versus limb muscle function in an experimental critical illness model.

Acute quadriplegic myopathy (AQM) is a common debilitating acquired disorder in critically ill intensive care unit (ICU) patients that is characterized by tetraplegia/generalized weakness of limb and trunk muscles. Masticatory muscles, on the other hand, are typically spared or less affected, yet the mechanisms underlying this striking muscle-specific difference remain unknown. This study aims to evaluate physiological parameters and the gene expression profiles of masticatory and limb muscles exposed to factors suggested to trigger AQM, such as mechanical ventilation, immobilization, neuromuscular blocking agents, corticosteroids (CS), and sepsis for 5 days by using a unique porcine model mimicking the ICU conditions. Single muscle fiber cross-sectional area and force-generating capacity, i.e., maximum force normalized to fiber cross-sectional area (specific force), revealed maintained masseter single muscle fiber cross-sectional area and specific-force after 5 days' exposure to all triggering factors. This is in sharp contrast to observations in limb and trunk muscles, showing a dramatic decline in specific force in response to 5 days' exposure to the triggering factors. Significant differences in gene expression were observed between craniofacial and limb muscles, indicating a highly complex and muscle-specific response involving transcription and growth factors, heat shock proteins, matrix metalloproteinase inhibitor, oxidative stress responsive elements, and sarcomeric proteins underlying the relative sparing of cranial vs. spinal nerve innervated muscles during exposure to the ICU intervention.

Muscle wasting and the temporal gene expression pattern in a novel rat intensive care unit model.

BACKGROUND: Acute quadriplegic myopathy (AQM) or critical illness myopathy (CIM) is frequently observed in intensive care unit (ICU) patients. To elucidate duration-dependent effects of the ICU intervention on molecular and functional networks that control the muscle wasting and weakness associated with AQM, a gene expression profile was analyzed at time points varying from 6 hours to 14 days in a unique experimental rat model mimicking ICU conditions, i.e., post-synaptically paralyzed, mechanically ventilated and extensively monitored animals. RESULTS: During the observation period, 1583 genes were significantly up- or down-regulated by factors of two or greater. A significant temporal gene expression pattern was constructed at short (6 h-4 days), intermediate (5-8 days) and long (9-14 days) durations. A striking early and maintained up-regulation (6 h-14d) of muscle atrogenes (muscle ring-finger 1/tripartite motif-containing 63 and F-box protein 32/atrogin-1) was observed, followed by an up-regulation of the proteolytic systems at intermediate and long durations (5-14d). Oxidative stress response genes and genes that take part in amino acid catabolism, cell cycle arrest, apoptosis, muscle development, and protein synthesis together with myogenic factors were significantly up-regulated from 5 to 14 days. At 9-14 d, genes involved in immune response and the caspase cascade were up-regulated. At 5-14d, genes related to contractile (myosin heavy chain and myosin binding protein C), regulatory (troponin, tropomyosin), developmental, caveolin-3, extracellular matrix, glycolysis/gluconeogenesis, cytoskeleton/sarcomere regulation and mitochondrial proteins were down-regulated. An activation of genes related to muscle growth and new muscle fiber formation (increase of myogenic factors and JunB and down-regulation of myostatin) and up-regulation of genes that code protein synthesis and translation factors were found from 5 to 14 days. CONCLUSIONS: Novel temporal patterns of gene expression have been uncovered, suggesting a unique, coordinated and highly complex mechanism underlying the muscle wasting associated with AQM in ICU patients and providing new target genes and avenues for intervention studies.

Array-based genomic screening at diagnosis and during follow-up in chronic lymphocytic leukemia.

BACKGROUND: High-resolution genomic microarrays enable simultaneous detection of copy-number aberrations such as the known recurrent aberrations in chronic lymphocytic leukemia [del(11q), del(13q), del(17p) and trisomy 12], and copy-number neutral loss of heterozygosity. Moreover, comparison of genomic profiles from sequential patients' samples allows detection of clonal evolution. DESIGN AND METHODS: We screened samples from 369 patients with newly diagnosed chronic lymphocytic leukemia from a population-based cohort using 250K single nucleotide polymorphism-arrays. Clonal evolution was evaluated in 59 follow-up samples obtained after 5-9 years. RESULTS: At diagnosis, copy-number aberrations were identified in 90% of patients; 70% carried known recurrent alterations, including del(13q) (55%), trisomy 12 (10.5%), del(11q) (10%), and del(17p) (4%). Additional recurrent aberrations were detected on chromosomes 2 (1.9%), 4 (1.4%), 8 (1.6%) and 14 (1.6%). Thirteen patients (3.5%) displayed recurrent copy-number neutral loss of heterozygosity on 13q, of whom 11 had concurrent homozygous del(13q). Genomic complexity and large 13q deletions correlated with inferior outcome, while the former was linked to poor-prognostic aberrations. In the follow-up study, clonal evolution developed in 8/24 (33%) patients with unmutated IGHV, and in 4/25 (16%) IGHV-mutated and treated patients. In contrast, untreated patients with mutated IGHV (n=10) did not acquire additional aberrations. The most common secondary event, del(13q), was detected in 6/12 (50%) of all patients with acquired alterations. Interestingly, aberrations on, for example, chromosome 6q, 8p, 9p and 10q developed exclusively in patients with unmutated IGHV. CONCLUSIONS: Whole-genome screening revealed a high frequency of genomic aberrations in newly diagnosed chronic lymphocytic leukemia. Clonal evolution was associated with other markers of aggressive disease and commonly included the known recurrent aberrations.

Distinct gene expression profiles in subsets of chronic lymphocytic leukemia expressing stereotyped IGHV4-34 B-cell receptors.

BACKGROUND: Numerous subsets of patients with chronic lymphocytic leukemia display similar immunoglobulin gene usage with almost identical complementarity determining region 3 sequences. Among IGHV4-34 cases, two such subsets with "stereotyped" B-cell receptors were recently identified, i.e. subset #4 (IGHV4-34/IGKV2-30) and subset #16 (IGHV4-34/IGKV3-20). Subset #4 patients appear to share biological and clinical features, e.g. young age at diagnosis and indolent disease, whereas little is known about subset #16 at a clinical level. DESIGN AND METHODS: We investigated the global gene expression pattern in sorted chronic lymphocytic leukemia cells from 25 subset/non-subset IGHV4-34 patients using Affymetrix gene expression arrays. RESULTS: Although generally few differences were found when comparing subset to non-subset 4/16 IGHV4-34 cases, distinct gene expression profiles were revealed for subset #4 versus subset #16. The differentially expressed genes, predominantly with lower expression in subset #4 patients, are involved in important cell regulatory pathways including cell-cycle control, proliferation and immune response, which may partly explain the low-proliferative disease observed in subset #4 patients. CONCLUSIONS: Our novel data demonstrate distinct gene expression profiles among patients with stereotyped IGHV4-34 B-cell receptors, providing further evidence for biological differences in the pathogenesis of these subsets and underscoring the functional relevance of subset assignment based on B-cell receptor sequence features.

High-density screening reveals a different spectrum of genomic aberrations in chronic lymphocytic leukemia patients with 'stereotyped' IGHV3-21 and IGHV4-34 B-cell receptors.

BACKGROUND: The existence of multiple subsets of chronic lymphocytic leukemia expressing 'stereotyped' B-cell receptors implies the involvement of antigen(s) in leukemogenesis. Studies also indicate that 'stereotypy' may influence the clinical course of patients with chronic lymphocytic leukemia, for example, in subsets with stereotyped IGHV3-21 and IGHV4-34 B-cell receptors; however, little is known regarding the genomic profile of patients in these subsets. DESIGN AND METHODS: We applied 250K single nucleotide polymorphism-arrays to study copy-number aberrations and copy-number neutral loss-of-heterozygosity in patients with stereotyped IGHV3-21 (subset #2, n=29), stereotyped IGHV4-34 (subset #4, n=17; subset #16, n=8) and non-subset #2 IGHV3-21 (n=13) and non-subset #4/16 IGHV4-34 (n=34) patients. RESULTS: Over 90% of patients in subset #2 and non-subset #2 carried copy-number aberrations, whereas 75-76% of patients in subset #4 and subset #16 showed copy-number aberrations. Subset #2 and non-subset #2 patients also displayed a higher average number of aberrations compared to patients in subset #4. Deletion of 13q was the only known recurrent aberration detected in subset #4 (35%); this aberration was even more frequent in subset #2 (79%). del(11q) was more frequent in subset #2 and non-subset #2 (31% and 23%) patients than in subset #4 and non-subset #4/16 patients. Recurrent copy-number neutral loss-of-heterozygosity was mainly detected on chromosome 13q, independently of B-cell receptor stereotypy. CONCLUSIONS: Genomic aberrations were more common in subset #2 and non-subset #2 than in subset #4. The particularly high frequency of del(11q) in subset #2 may be linked to the adverse outcome reported for patients in this subset. Conversely, the lower prevalence of copy-number aberrations and the absence of poor-prognostic aberrations in subset #4 may reflect an inherently low-proliferative disease, which would prevent accumulation of genomic alterations.

Gene expression and muscle fiber function in a porcine ICU model.

Skeletal muscle wasting and impaired muscle function in response to mechanical ventilation and immobilization in intensive care unit (ICU) patients are clinically challenging partly due to 1) the poorly understood intricate cellular and molecular networks and 2) the unavailability of an animal model mimicking this condition. By employing a unique porcine model mimicking the conditions in the ICU with long-term mechanical ventilation and immobilization, we have analyzed the expression profile of skeletal muscle biopsies taken at three time points during a 5-day period. Among the differentially regulated transcripts, extracellular matrix, energy metabolism, sarcomeric and LIM protein mRNA levels were downregulated, while ubiquitin proteasome system, cathepsins, oxidative stress responsive genes and heat shock proteins (HSP) mRNAs were upregulated. Despite 5 days of immobilization and mechanical ventilation single muscle fiber cross-sectional areas as well as the maximum force generating capacity at the single muscle fiber level were preserved. It is proposed that HSP induction in skeletal muscle is an inherent, primary, but temporary protective mechanism against protein degradation. To our knowledge, this is the first study that isolates the effect of immobilization and mechanical ventilation in an ICU condition from various other cofactors.

Extensive expression of craniofacial related homeobox genes in canine mammary sarcomas.

The global gene expression in three types of canine mammary tumors: carcinoma, fibrosarcoma and osteosarcoma were investigated by Affymetrix gene array technology. Unsupervised clustering analysis revealed a close clustering of the respective tumor types, with fibrosarcomas clustering close to the osteosarcomas and the carcinomas clustering closer to non-malignant mammary tissues (NMTs). A number of epithelial markers were expressed in both carcinomas and NMTs, whereas the sarcomas expressed genes related to mesenchymal differentiation. A comparison of the gene expression profile of the sarcomas versus carcinoma/NMTs revealed that the sarcomas, in particular the osteosarcomas, showed a striking upregulation of a panel of homeobox genes previously linked to craniofacial bone formation. In line with this finding, osteosarcomas showed an upregulation of bone morphogenetic proteins (BMPs) and of genes associated with retinoic acid signaling. Increased homeobox gene expression in sarcomas was also confirmed at the protein level by immunohistochemical analysis of tumor tissue, and in an osteosarcoma cell line after stimulation by BMP-2. These findings suggest that the development of mammary sarcomas specifically involves triggering of a set of homeobox genes related to neural crest and craniofacial bone development.

Screening for copy-number alterations and loss of heterozygosity in chronic lymphocytic leukemia--a comparative study of four differently designed, high resolution microarray platforms.

Screening for gene copy-number alterations (CNAs) has improved by applying genome-wide microarrays, where SNP arrays also allow analysis of loss of heterozygozity (LOH). We here analyzed 10 chronic lymphocytic leukemia (CLL) samples using four different high-resolution platforms: BAC arrays (32K), oligonucleotide arrays (185K, Agilent), and two SNP arrays (250K, Affymetrix and 317K, Illumina). Cross-platform comparison revealed 29 concordantly detected CNAs, including known recurrent alterations, which confirmed that all platforms are powerful tools when screening for large aberrations. However, detection of 32 additional regions present in 2-3 platforms illustrated a discrepancy in detection of small CNAs, which often involved reported copy-number variations. LOH analysis using dChip revealed concordance of mainly large regions, but showed numerous, small nonoverlapping regions and LOH escaping detection. Evaluation of baseline variation and copy-number ratio response showed the best performance for the Agilent platform and confirmed the robustness of BAC arrays. Accordingly, these platforms demonstrated a higher degree of platform-specific CNAs. The SNP arrays displayed higher technical variation, although this was compensated by high density of elements. Affymetrix detected a higher degree of CNAs compared to Illumina, while the latter showed a lower noise level and higher detection rate in the LOH analysis. Large-scale studies of genomic aberrations are now feasible, but new tools for LOH analysis are requested.

Gene expression analysis of kidneys from transgenic mice expressing fibroblast growth factor-23.

BACKGROUND: Fibroblast growth factor-23 (FGF23), a circulating protein produced in bone, causes decreased renal inorganic phosphate (Pi) reabsorption by reducing the expression of the sodium phosphate cotransporter type 2a (Npt2a). We have previously generated transgenic mice expressing human wild-type (WT) FGF23 under the control of the alpha1 (I) collagen promoter. METHODS: In this study, we performed a large-scale gene expression study of kidneys from FGF23 transgenic mice and WT littermates. Microarray expression data of key transcripts were verified by real-time RT-PCR analysis. RESULTS: Several genes that play a role in Pi regulation revealed decreased expression levels in the transgenic mice, such as Npt2a and Pdzk1, a scaffolding protein known to interact with Npt2a. Importantly, Klotho, a suggested FGF23 receptor cofactor, was the most significantly decreased transcript and alpha2-Na(+)/K(+)-ATPase (Atp1a2), a gene isoform of alpha1-Na(+)/K(+)-ATPase (Atp1a1) which has recently been shown to interact with Klotho and regulate calcium metabolism, was the most increased transcript. In contrast, other genes proposed to regulate Pi levels, such as secreted frizzled-related protein-4 (sFrp4) and Na(+)/H(+) exchanger regulatory factor-1 (Nherf1) revealed no changes. CONCLUSIONS: FGF23 transgenic mice display differentially expressed transcript levels of several genes essential in renal Pi regulation. These findings may lead to further understanding of how FGF23 mediates its actions on renal Pi regulation.

Gene expression analysis identifies a genetic signature potentially associated with response to alpha-IFN in chronic phase CML patients.

Microarray-based gene expression analysis was performed on diagnostic chronic phase CML patient samples prior to interferon treatment. Fifteen patient samples corresponding to six cytogenetic responders and nine non-responders were included. Genes differentially expressed between responder and non-responder patients were listed and a subsequent leave-one-out cross validation (LOOV) procedure showed that the top 20 genes allowed the highest prediction accuracy. The relevant genes were quantified by real-time PCR that supported the microarray results. We conclude that it might be possible to use gene expression analysis to predict future response to interferon in CML diagnostic samples.

Expression analysis of genes involved in brain tumor progression driven by retroviral insertional mutagenesis in mice.

Retroviral tagging previously identified putative cancer-causing genes in a mouse brain tumor model where a recombinant Moloney murine leukemia virus encoding the platelet-derived growth factor B-chain (MMLV/PDGFB) was intracerebrally injected in newborn mice. In the present study, expression analysis using cDNA arrays revealed several similarities of virus-induced mouse gliomas with human brain tumors. Brain tumors with short latency contained on average 8.0 retroviral insertions and resembled human glioblastoma multiforme (GBM) whereas long-latency gliomas were of lower grade, similar to human oligodendroglioma (OD) and had 2.3 insertions per tumor. Several known and novel genes of tumor progression or cell markers were differentially expressed between OD- and GBM-like tumors. Array and quantitative real-time PCR analysis demonstrated elevated expression similar to Pdgfralpha of retrovirally tagged genes Abhd2, Ddr1, Fos, Ng2, Ppfibp1, Rad51b and Sulf2 in both glioma types compared to neonatal and adult normal brain. The retrovirally tagged genes Plekhb1, Prex1, Prkg2, Sox10 and 1200004M23Rik were upregulated in the tumors but had a different expression profile than Pdgfralpha whereas Rap1gap, Gli1, Neurl and Camk2b were downregulated in the tumors. The present study accentuates the proposed role of the retrovirally tagged genes in PDGF-driven gliomagenesis and indicates that insertional mutagenesis can promote glioma progression.

Improved variance estimation of classification performance via reduction of bias caused by small sample size.

BACKGROUND: Supervised learning for classification of cancer employs a set of design examples to learn how to discriminate between tumors. In practice it is crucial to confirm that the classifier is robust with good generalization performance to new examples, or at least that it performs better than random guessing. A suggested alternative is to obtain a confidence interval of the error rate using repeated design and test sets selected from available examples. However, it is known that even in the ideal situation of repeated designs and tests with completely novel samples in each cycle, a small test set size leads to a large bias in the estimate of the true variance between design sets. Therefore different methods for small sample performance estimation such as a recently proposed procedure called Repeated Random Sampling (RSS) is also expected to result in heavily biased estimates, which in turn translates into biased confidence intervals. Here we explore such biases and develop a refined algorithm called Repeated Independent Design and Test (RIDT). RESULTS: Our simulations reveal that repeated designs and tests based on resampling in a fixed bag of samples yield a biased variance estimate. We also demonstrate that it is possible to obtain an improved variance estimate by means of a procedure that explicitly models how this bias depends on the number of samples used for testing. For the special case of repeated designs and tests using new samples for each design and test, we present an exact analytical expression for how the expected value of the bias decreases with the size of the test set. CONCLUSION: We show that via modeling and subsequent reduction of the small sample bias, it is possible to obtain an improved estimate of the variance of classifier performance between design sets. However, the uncertainty of the variance estimate is large in the simulations performed indicating that the method in its present form cannot be directly applied to small data sets.

Differential Gene Regulation in Fibroblasts in Co-culture with Keratinocytes and Head and Neck SCC Cells.

BACKGROUND: While carcinoma-associated fibroblasts (CAFs) support tumorigenesis, normal tissue fibroblasts suppress tumor progression. Mechanisms behind conversion of fibroblasts into a CAF phenotype are largely unrevealed. MATERIALS AND METHODS: Transwell co-cultures with fibroblasts in collagen gels and squamous-cell carcinoma (SCC) cells or normal oral keratinocytes (NOKs) in inserts. Differences in fibroblast global gene expression were analyzed using Affymetrix arrays and subsequent functional annotation and cluster analysis, as well as gene set enrichment analysis were performed. RESULTS: There were 52 up-regulated and 30 down-regulated transcript IDs (>2-fold, p<0.05) in fibroblasts co-cultured with SCC compared to NOKs. Functional analysis demonstrated an enrichment of collagen-related genes. There were similarities with gene sets reflecting a non-specific, innate-type response with activation of both interferon pathways and connective tissue turnover. CONCLUSION: There were distinct differences in fibroblast gene expression between the co-culture types. Many were in genes related to an innate-type of response and to connective tissue turnover.

Alpha terpineol: a potential anticancer agent which acts through suppressing NF-kappaB signalling.

BACKGROUND: Alpha terpineol is a bioactive component of Salvia libanotica essential oil extract and has shown antitumour activity. MATERIALS AND METHODS: The cytotoxicity of alpha terpineol towards different tumour cell lines was evaluated in vitro. Mechanistic characterization was performed using analysis of drug activity in a cell line panel and drug-induced gene expression perturbation using the connectivity map approach. RESULTS: The small cell lung carcinoma was the cell line most sensitive to alpha terpineol. The results proposed alpha terpineol as an NF-kappaB inhibitor, which was confirmed by the observed dose-dependent inhibition of NF-kappaB translocation and activity using two NF-kappaB assays, and by the down-regulation of the expression of several NF-kappaB-related genes such as IL-1 beta and IL1R1. CONCLUSION: The results suggest that alpha terpineol inhibits the growth of tumour cells through a mechanism that involves inhibition of the NF-kappaB pathway.

Improving Bayesian credibility intervals for classifier error rates using maximum entropy empirical priors.

OBJECTIVE: Successful use of classifiers that learn to make decisions from a set of patient examples require robust methods for performance estimation. Recently many promising approaches for determination of an upper bound for the error rate of a single classifier have been reported but the Bayesian credibility interval (CI) obtained from a conventional holdout test still delivers one of the tightest bounds. The conventional Bayesian CI becomes unacceptably large in real world applications where the test set sizes are less than a few hundred. The source of this problem is that fact that the CI is determined exclusively by the result on the test examples. In other words, there is no information at all provided by the uniform prior density distribution employed which reflects complete lack of prior knowledge about the unknown error rate. Therefore, the aim of the study reported here was to study a maximum entropy (ME) based approach to improved prior knowledge and Bayesian CIs, demonstrating its relevance for biomedical research and clinical practice. METHOD AND MATERIAL: It is demonstrated how a refined non-uniform prior density distribution can be obtained by means of the ME principle using empirical results from a few designs and tests using non-overlapping sets of examples. RESULTS: Experimental results show that ME based priors improve the CIs when employed to four quite different simulated and two real world data sets. CONCLUSIONS: An empirically derived ME prior seems promising for improving the Bayesian CI for the unknown error rate of a designed classifier.

Quantification of normal cell fraction and copy number neutral LOH in clinical lung cancer samples using SNP array data.

BACKGROUND: Technologies based on DNA microarrays have the potential to provide detailed information on genomic aberrations in tumor cells. In practice a major obstacle for quantitative detection of aberrations is the heterogeneity of clinical tumor tissue. Since tumor tissue invariably contains genetically normal stromal cells, this may lead to a failure to detect aberrations in the tumor cells. PRINCIPAL FINDING: Using SNP array data from 44 non-small cell lung cancer samples we have developed a bioinformatic algorithm that accurately models the fractions of normal and tumor cells in clinical tumor samples. The proportion of normal cells in combination with SNP array data can be used to detect and quantify copy number neutral loss-of-heterozygosity (CNNLOH) in the tumor cells both in crude tumor tissue and in samples enriched for tumor cells by laser capture microdissection. CONCLUSION: Genome-wide quantitative analysis of CNNLOH using the CNNLOH Quantifier method can help to identify recurrent aberrations contributing to tumor development in clinical tumor samples. In addition, SNP-array based analysis of CNNLOH may become important for detection of aberrations that can be used for diagnostic and prognostic purposes.

Segmentation-based detection of allelic imbalance and loss-of-heterozygosity in cancer cells using whole genome SNP arrays.

We present a strategy for detection of loss-of-heterozygosity and allelic imbalance in cancer cells from whole genome single nucleotide polymorphism genotyping data. Using a dilution series of a tumor cell line mixed with its paired normal cell line and data generated on Affymetrix and Illumina platforms, including paired tumor-normal samples and tumors characterized by fluorescent in situ hybridization, we demonstrate a high sensitivity and specificity of the strategy for detecting both minute and gross allelic imbalances in heterogeneous tumor samples.