Semimethylation is a feature of diffuse large B-cell lymphoma, and subgroups with poor prognosis are characterized by global hypomethylation and short telomere length.

BACKGROUND: Large B-cell lymphoma (LBCL) is the most common lymphoma and is known to be a biologically heterogeneous disease regarding genetic, phenotypic, and clinical features. Although the prognosis is good, one-third has a primary refractory or relapsing disease which underscores the importance of developing predictive biological markers capable of identifying high- and low-risk patients. DNA methylation (DNAm) and telomere maintenance alterations are hallmarks of cancer and aging. Both these alterations may contribute to the heterogeneity of the disease, and potentially influence the prognosis of LBCL. RESULTS: We studied the DNAm profiles (Infinium MethylationEPIC BeadChip) and relative telomere lengths (RTL) with qPCR of 93 LBCL cases: Diffuse large B-cell lymphoma not otherwise specified (DLBCL, n = 66), High-grade B-cell lymphoma (n = 7), Primary CNS lymphoma (n = 8), and transformation of indolent B-cell lymphoma (n = 12). There was a substantial methylation heterogeneity in DLBCL and other LBCL entities compared to normal cells and other B-cell neoplasms. LBCL cases had a particularly aberrant semimethylated pattern (0.15 ≤ ß ≤ 0.8) with large intertumor variation and overall low hypermethylation (ß > 0.8). DNAm patterns could not be used to distinguish between germinal center B-cell-like (GC) and non-GC DLBCL cases. In cases treated with R-CHOP-like regimens, a high percentage of global hypomethylation (ß < 0.15) was in multivariable analysis associated with worse disease-specific survival (DSS) (HR 6.920, 95% CI 1.499-31.943) and progression-free survival (PFS) (HR 4.923, 95% CI 1.286-18.849) in DLBCL and with worse DSS (HR 5.147, 95% CI 1.239-21.388) in LBCL. These cases with a high percentage of global hypomethylation also had a higher degree of CpG island methylation, including islands in promoter-associated regions, than the cases with less hypomethylation. Additionally, telomere length was heterogenous in LBCL, with a subset of the DLBCL-GC cases accounting for the longest RTL. Short RTL was independently associated with worse DSS (HR 6.011, 95% CI 1.319-27.397) and PFS (HR 4.689, 95% CI 1.102-19.963) in LBCL treated with R-CHOP-like regimens. CONCLUSION: We hypothesize that subclones with high global hypomethylation and hypermethylated CpG islands could have advantages in tumor progression, e.g. by inactivating tumor suppressor genes or promoting treatment resistance. Our findings suggest that cases with high global hypomethylation and thus poor prognosis could be candidates for alternative treatment regimens including hypomethylating drugs.

Pre-Clinical Evaluation of the Hypomethylating Agent Decitabine for the Treatment of T-Cell Lymphoblastic Lymphoma.

T-cell lymphoblastic lymphoma (T-LBL) is a rare and aggressive lymphatic cancer, often diagnosed at a young age. Patients are treated with intensive chemotherapy, potentially followed by a hematopoietic stem cell transplantation. Although prognosis of T-LBL has improved with intensified treatment protocols, they are associated with side effects and 10-20% of patients still die from relapsed or refractory disease. Given this, the search toward less toxic anti-lymphoma therapies is ongoing. Here, we targeted the recently described DNA hypermethylated profile in T-LBL with the DNA hypomethylating agent decitabine. We evaluated the anti-lymphoma properties and downstream effects of decitabine, using patient derived xenograft (PDX) models. Decitabine treatment resulted in prolonged lymphoma-free survival in all T-LBL PDX models, which was associated with downregulation of the oncogenic MYC pathway. However, some PDX models showed more benefit of decitabine treatment compared to others. In more sensitive models, differentially methylated CpG regions resulted in more differentially expressed genes in open chromatin regions. This resulted in stronger downregulation of cell cycle genes and upregulation of immune response activating transcripts. Finally, we suggest a gene signature for high decitabine sensitivity in T-LBL. Altogether, we here delivered pre-clinical proof of the potential use of decitabine as a new therapeutic agent in T-LBL.

A novel DNA methylation signature is associated with androgen receptor activity and patient prognosis in bone metastatic prostate cancer.

BACKGROUND: Patients with metastatic prostate cancer (PC) are treated with androgen deprivation therapy (ADT) that initially reduces metastasis growth, but after some time lethal castration-resistant PC (CRPC) develops. A better understanding of the tumor biology in bone metastases is needed to guide further treatment developments. Subgroups of PC bone metastases based on transcriptome profiling have been previously identified by our research team, and specifically, heterogeneities related to androgen receptor (AR) activity have been described. Epigenetic alterations during PC progression remain elusive and this study aims to explore promoter gene methylation signatures in relation to gene expression and tumor AR activity. MATERIALS AND METHODS: Genome-wide promoter-associated CpG methylation signatures of a total of 94 tumor samples, including paired non-malignant and malignant primary tumor areas originating from radical prostatectomy samples (n = 12), and bone metastasis samples of separate patients with hormone-naive (n = 14), short-term castrated (n = 4) or CRPC (n = 52) disease were analyzed using the Infinium Methylation EPIC arrays, along with gene expression analysis by Illumina Bead Chip arrays (n = 90). AR activity was defined from expression levels of genes associated with canonical AR activity. RESULTS: Integrated epigenome and transcriptome analysis identified pronounced hypermethylation in malignant compared to non-malignant areas of localized prostate tumors. Metastases showed an overall hypomethylation in relation to primary PC, including CpGs in the AR promoter accompanied with induction of AR mRNA levels. We identified a Methylation Classifier for Androgen receptor activity (MCA) signature, which separated metastases into two clusters (MCA positive/negative) related to tumor characteristics and patient prognosis. The MCA positive metastases showed low methylation levels of genes associated with canonical AR signaling and patients had a more favorable prognosis after ADT. In contrast, MCA negative patients had low AR activity associated with hypermethylation of AR-associated genes, and a worse prognosis after ADT. CONCLUSIONS: A promoter methylation signature classifies PC bone metastases into two groups and predicts tumor AR activity and patient prognosis after ADT. The explanation for the methylation diversities observed during PC progression and their biological and clinical relevance need further exploration.

Combining epigenetic and clinicopathological variables improves specificity in prognostic prediction in clear cell renal cell carcinoma.

BACKGROUND: Metastasized clear cell renal cell carcinoma (ccRCC) is associated with a poor prognosis. Almost one-third of patients with non-metastatic tumors at diagnosis will later progress with metastatic disease. These patients need to be identified already at diagnosis, to undertake closer follow up and/or adjuvant treatment. Today, clinicopathological variables are used to risk classify patients, but molecular biomarkers are needed to improve risk classification to identify the high-risk patients which will benefit most from modern adjuvant therapies. Interestingly, DNA methylation profiling has emerged as a promising prognostic biomarker in ccRCC. This study aimed to derive a model for prediction of tumor progression after nephrectomy in non-metastatic ccRCC by combining DNA methylation profiling with clinicopathological variables. METHODS: A novel cluster analysis approach (Directed Cluster Analysis) was used to identify molecular biomarkers from genome-wide methylation array data. These novel DNA methylation biomarkers, together with previously identified CpG-site biomarkers and clinicopathological variables, were used to derive predictive classifiers for tumor progression. RESULTS: The "triple classifier" which included both novel and previously identified DNA methylation biomarkers together with clinicopathological variables predicted tumor progression more accurately than the currently used Mayo scoring system, by increasing the specificity from 50% in Mayo to 64% in our triple classifier at 85% fixed sensitivity. The cumulative incidence of progress (pCIP5yr) was 7.5% in low-risk vs 44.7% in high-risk in M0 patients classified by the triple classifier at diagnosis. CONCLUSIONS: The triple classifier panel that combines clinicopathological variables with genome-wide methylation data has the potential to improve specificity in prognosis prediction for patients with non-metastatic ccRCC.

Aging of preleukemic thymocytes drives CpG island hypermethylation in T-cell acute lymphoblastic leukemia.

Cancer cells display DNA hypermethylation at specific CpG islands in comparison to their normal healthy counterparts, but the mechanism that drives this so-called CpG island methylator phenotype (CIMP) remains poorly understood. Here, we show that CpG island methylation in human T-cell acute lymphoblastic leukemia (T-ALL) mainly occurs at promoters of Polycomb Repressor Complex 2 (PRC2) target genes that are not expressed in normal or malignant T-cells and which display a reciprocal association with H3K27me3 binding. In addition, we revealed that this aberrant methylation profile reflects the epigenetic history of T-ALL and is established already in pre-leukemic, self-renewing thymocytes that precede T-ALL development. Finally, we unexpectedly uncover that this age-related CpG island hypermethylation signature in T-ALL is completely resistant to the FDA-approved hypomethylating agent Decitabine. Altogether, we here provide conceptual evidence for the involvement of a pre-leukemic phase characterized by self-renewing thymocytes in the pathogenesis of human T-ALL.

DNA methylation and copy number variation profiling of T-cell lymphoblastic leukemia and lymphoma.

Despite having common overlapping immunophenotypic and morphological features, T-cell lymphoblastic leukemia (T-ALL) and lymphoma (T-LBL) have distinct clinical manifestations, which may represent separate diseases. We investigated and compared the epigenetic and genetic landscape of adult and pediatric T-ALL (n = 77) and T-LBL (n = 15) patient samples by high-resolution genome-wide DNA methylation and Copy Number Variation (CNV) BeadChip arrays. DNA methylation profiling identified the presence of CpG island methylator phenotype (CIMP) subgroups within both pediatric and adult T-LBL and T-ALL. An epigenetic signature of 128 differentially methylated CpG sites was identified, that clustered T-LBL and T-ALL separately. The most significant differentially methylated gene loci included the SGCE/PEG10 shared promoter region, previously implicated in lymphoid malignancies. CNV analysis confirmed overlapping recurrent aberrations between T-ALL and T-LBL, including 9p21.3 (CDKN2A/CDKN2B) deletions. A significantly higher frequency of chromosome 13q14.2 deletions was identified in T-LBL samples (36% in T-LBL vs. 0% in T-ALL). This deletion, encompassing the RB1, MIR15A and MIR16-1 gene loci, has been reported as a recurrent deletion in B-cell malignancies. Our study reveals epigenetic and genetic markers that can distinguish between T-LBL and T-ALL, and deepen the understanding of the biology underlying the diverse disease localization.

Hematopoietic cellular aging is not accelerated during the first 2 years of life in children born preterm.

BACKGROUND: Prematurity in itself and exposure to neonatal intensive care triggers inflammatory processes and oxidative stress, leading to risk for disease later in life. The effects on cellular aging processes are incompletely understood. METHODS: Relative telomere length (RTL) was measured by qPCR in this longitudinal cohort study with blood samples taken at birth and at 2 years of age from 60 children (16 preterm and 44 term). Viral respiratory infections the first year were evaluated. Epigenetic biological DNA methylation (DNAm) age was predicted based on methylation array data in 23 children (11 preterm and 12 term). RTL change/year and DNAm age change/year was compared in preterm and term during the 2 first years of life. RESULTS: Preterm infants had longer telomeres than term born at birth and at 2 years of age, but no difference in telomere attrition rate could be detected. Predicted epigenetic DNAm age was younger in preterm infants, but rate of DNAm aging was similar in both groups. CONCLUSIONS: Despite early exposure to risk factors for accelerated cellular aging, children born preterm exhibited preserved telomeres. Stress during the neonatal intensive care period did not reflect accelerated epigenetic DNAm aging. Early-life aging was not explained by preterm birth. IMPACT: Preterm birth is associated with elevated disease risk later in life. Preterm children often suffer from inflammation early in life. Stress-related telomere erosion during neonatal intensive care has been proposed. Inflammation-accelerated biological aging in preterm is unknown. We find no accelerated aging due to prematurity or infections during the first 2 years of life.

DNA methylation associates with survival in non-metastatic clear cell renal cell carcinoma.

BACKGROUND: Clear cell renal cell carcinoma (ccRCC) is the most common subtype among renal cancer and is associated with poor prognosis if metastasized. Up to one third of patients with local disease at diagnosis will develop metastasis after nephrectomy, and there is a need for new molecular markers to identify patients with high risk of tumor progression. In the present study, we performed genome-wide promoter DNA methylation analysis at diagnosis to identify DNA methylation profiles associated with risk for progress. METHOD: Diagnostic tissue samples from 115 ccRCC patients were analysed by Illumina HumanMethylation450K arrays and methylation status of 155,931 promoter associated CpGs were related to genetic aberrations, gene expression and clinicopathological parameters. RESULTS: The ccRCC samples separated into two clusters (cluster A/B) based on genome-wide promoter methylation status. The samples in these clusters differed in tumor diameter (p < 0.001), TNM stage (p < 0.001), morphological grade (p < 0.001), and patients outcome (5 year cancer specific survival (pCSS5yr) p < 0.001 and cumulative incidence of progress (pCIP5yr) p < 0.001. An integrated genomic and epigenomic analysis in the ccRCCs, revealed significant correlations between the total number of genetic aberrations and total number of hypermethylated CpGs (R = 0.435, p < 0.001), and predicted mitotic age (R = 0.407, p < 0.001). We identified a promoter methylation classifier (PMC) panel consisting of 172 differently methylated CpGs accompanying progress of disease. Classifying non-metastatic patients using the PMC panel showed that PMC high tumors had a worse prognosis compared with the PMC low tumors (pCIP5yr 38% vs. 8%, p = 0.001), which was confirmed in non-metastatic ccRCCs in the publically available TCGA-KIRC dataset (pCIP5yr 39% vs. 16%, p < 0.001). CONCLUSION: DNA methylation analysis at diagnosis in ccRCC has the potential to improve outcome-prediction in non-metastatic patients at diagnosis.

An integrated transcriptome analysis in T-cell acute lymphoblastic leukemia links DNA methylation subgroups to dysregulated TAL1 and ANTP homeobox gene expression.

Classification of pediatric T-cell acute lymphoblastic leukemia (T-ALL) patients into CIMP (CpG Island Methylator Phenotype) subgroups has the potential to improve current risk stratification. To investigate the biology behind these CIMP subgroups, diagnostic samples from Nordic pediatric T-ALL patients were characterized by genome-wide methylation arrays, followed by targeted exome sequencing, telomere length measurement, and RNA sequencing. The CIMP subgroups did not correlate significantly with variations in epigenetic regulators. However, the CIMP+ subgroup, associated with better prognosis, showed indicators of longer replicative history, including shorter telomere length (P = 0.015) and older epigenetic (P < 0.001) and mitotic age (P < 0.001). Moreover, the CIMP+ subgroup had significantly higher expression of ANTP homeobox oncogenes, namely TLX3, HOXA9, HOXA10, and NKX2-1, and novel genes in T-ALL biology including PLCB4, PLXND1, and MYO18B. The CIMP- subgroup, with worse prognosis, was associated with higher expression of TAL1 along with frequent STIL-TAL1 fusions (2/40 in CIMP+ vs 11/24 in CIMP-), as well as stronger expression of BEX1. Altogether, our findings suggest different routes for leukemogenic transformation in the T-ALL CIMP subgroups, indicated by different replicative histories and distinct methylomic and transcriptomic profiles. These novel findings can lead to new therapeutic strategies.

DNA methylation holds prognostic information in relapsed precursor B-cell acute lymphoblastic leukemia.

Background: Few biological markers are associated with survival after relapse of B-cell precursor acute lymphoblastic leukemia (BCP-ALL). In pediatric T-cell ALL, we have identified promoter-associated methylation alterations that correlate with prognosis. Here, the prognostic relevance of CpG island methylation phenotype (CIMP) classification was investigated in pediatric BCP-ALL patients. Methods: Six hundred and one BCP-ALL samples from Nordic pediatric patients (age 1-18) were CIMP classified at initial diagnosis and analyzed in relation to clinical data. Results: Among the 137 patients that later relapsed, patients with a CIMP- profile (n = 42) at initial diagnosis had an inferior overall survival (pOS5years 33%) compared to CIMP+ patients (n = 95, pOS5years 65%) (p = 0.001), which remained significant in a Cox proportional hazards model including previously defined risk factors. Conclusion: CIMP classification is a strong candidate for improved risk stratification of relapsed BCP-ALL.

Decreased telomere length in children with cartilage-hair hypoplasia.

BACKGROUND: Cartilage-hair hypoplasia (CHH) is an autosomal recessive chondrodysplasia caused by RMRP (RNA component of mitochondrial RNA processing endoribonuclease) gene mutations. Manifestations include short stature, variable immunodeficiency, anaemia and increased risk of malignancies, all of which have been described also in telomere biology disorders. RMRP interacts with the telomerase RT (TERT) subunit, but the influence of RMRP mutations on telomere length is unknown. We measured relative telomere length (RTL) in patients with CHH, their first-degree relatives and healthy controls and correlated RTL with clinical and laboratory features. METHODS: The study cohort included 48 patients with CHH with homozygous (n=36) or compound heterozygous RMRP mutations (median age 38.2 years, range 6.0-70.8 years), 86 relatives (74 with a heterozygous RMRP mutation) and 94 unrelated healthy controls. We extracted DNA from peripheral blood, sequenced the RMRP gene and measured RTL by qPCR. RESULTS: Compared with age-matched and sex-matched healthy controls, median RTL was significantly shorter in patients with CHH (n=40 pairs, 1.05 vs 1.21, p=0.017), but not in mutation carriers (n=48 pairs, 1.16 vs 1.10, p=0.224). RTL correlated significantly with age in RMRP mutation carriers (r=-0.482, p<0.001) and non-carriers (r=-0.498, p<0.001), but not in patients (r=-0.236, p=0.107). In particular children (<18 years) with CHH had shorter telomeres than controls (median RTL 1.12 vs 1.26, p=0.008). In patients with CHH, RTL showed no correlation with genotype, clinical or laboratory characteristics. CONCLUSIONS: Telomere length was decreased in children with CHH. We found no correlation between RTL and clinical or laboratory parameters.

DNA Methylation Adds Prognostic Value to Minimal Residual Disease Status in Pediatric T-Cell Acute Lymphoblastic Leukemia.

BACKGROUND: Despite increased knowledge about genetic aberrations in pediatric T-cell acute lymphoblastic leukemia (T-ALL), no clinically feasible treatment-stratifying marker exists at diagnosis. Instead patients are enrolled in intensive induction therapies with substantial side effects. In modern protocols, therapy response is monitored by minimal residual disease (MRD) analysis and used for postinduction risk group stratification. DNA methylation profiling is a candidate for subtype discrimination at diagnosis and we investigated its role as a prognostic marker in pediatric T-ALL. PROCEDURE: Sixty-five diagnostic T-ALL samples from Nordic pediatric patients treated according to the Nordic Society of Pediatric Hematology and Oncology ALL 2008 (NOPHO ALL 2008) protocol were analyzed by HumMeth450K genome wide DNA methylation arrays. Methylation status was analyzed in relation to clinical data and early T-cell precursor (ETP) phenotype. RESULTS: Two distinct CpG island methylator phenotype (CIMP) groups were identified. Patients with a CIMP-negative profile had an inferior response to treatment compared to CIMP-positive patients (3-year cumulative incidence of relapse (CIR3y ) rate: 29% vs. 6%, P = 0.01). Most importantly, CIMP classification at diagnosis allowed subgrouping of high-risk T-ALL patients (MRD ≥0.1% at day 29) into two groups with significant differences in outcome (CIR3y rates: CIMP negative 50% vs. CIMP positive 12%; P = 0.02). These groups did not differ regarding ETP phenotype, but the CIMP-negative group was younger (P = 0.02) and had higher white blood cell count at diagnosis (P = 0.004) compared with the CIMP-positive group. CONCLUSIONS: CIMP classification at diagnosis in combination with MRD during induction therapy is a strong candidate for further risk classification and could confer important information in treatment decision making.

DNA methylation status defines clinicopathological parameters including survival for patients with clear cell renal cell carcinoma (ccRCC).

Epigenetic alterations in the methylome have been associated with tumor development and progression in renal cell carcinoma (RCC). In this study, 45 tumor samples, 12 tumor-free kidney cortex tissues, and 24 peripheral blood samples from patients with clear cell RCC (ccRCC) were analyzed by genome-wide promoter-directed methylation arrays and related to clinicopathological parameters. Unsupervised hierarchical clustering separated the tumors into two distinct methylation groups (clusters A and B), where cluster B had higher average methylation and increased number of hypermethylated CpG sites (CpGs). Furthermore, tumors in cluster B had, compared with cluster A, a larger tumor diameter (p = 0.033), a higher morphologic grade (p < 0.001), a higher tumor-node-metastasis (TNM) stage (p < 0.001), and a worse prognosis (p = 0.005). Higher TNM stage was correlated to an increase in average methylation level (p = 0.003) and number of hypermethylated CpGs (p = 0.003), whereas a number of hypomethylated CpGs were mainly unchanged. However, the predicted age of the tumors based on methylation profile did not correlate with TNM stage, morphological grade, or methylation cluster. Differently methylated (DM) genes (n = 840) in ccRCC samples compared with tumor-free kidney cortex samples were predominantly hypermethylated and a high proportion were identified as polycomb target genes. The DM genes were overrepresented by transcription factors, ligands, and receptors, indicating functional alterations of significance for ccRCC progression. To conclude, increased number of hypermethylated genes was associated with increased TNM stage of the tumors. DNA methylation classification of ccRCC tumor samples at diagnosis can serve as a clinically applicable prognostic marker in ccRCC.

Immortalization of T-cells is accompanied by gradual changes in CpG methylation resulting in a profile resembling a subset of T-cell leukemias.

We have previously described gene expression changes during spontaneous immortalization of T-cells, thereby identifying cellular processes important for cell growth crisis escape and unlimited proliferation. Here, we analyze the same model to investigate the role of genome-wide methylation in the immortalization process at different time points pre-crisis and post-crisis using high-resolution arrays. We show that over time in culture there is an overall accumulation of methylation alterations, with preferential increased methylation close to transcription start sites (TSSs), islands, and shore regions. Methylation and gene expression alterations did not correlate for the majority of genes, but for the fraction that correlated, gain of methylation close to TSS was associated with decreased gene expression. Interestingly, the pattern of CpG site methylation observed in immortal T-cell cultures was similar to clinical T-cell acute lymphoblastic leukemia (T-ALL) samples classified as CpG island methylator phenotype positive. These sites were highly overrepresented by polycomb target genes and involved in developmental, cell adhesion, and cell signaling processes. The presence of non-random methylation events in in vitro immortalized T-cell cultures and diagnostic T-ALL samples indicates altered methylation of CpG sites with a possible role in malignant hematopoiesis.

Normalization of high dimensional genomics data where the distribution of the altered variables is skewed.

Genome-wide analysis of gene expression or protein binding patterns using different array or sequencing based technologies is now routinely performed to compare different populations, such as treatment and reference groups. It is often necessary to normalize the data obtained to remove technical variation introduced in the course of conducting experimental work, but standard normalization techniques are not capable of eliminating technical bias in cases where the distribution of the truly altered variables is skewed, i.e. when a large fraction of the variables are either positively or negatively affected by the treatment. However, several experiments are likely to generate such skewed distributions, including ChIP-chip experiments for the study of chromatin, gene expression experiments for the study of apoptosis, and SNP-studies of copy number variation in normal and tumour tissues. A preliminary study using spike-in array data established that the capacity of an experiment to identify altered variables and generate unbiased estimates of the fold change decreases as the fraction of altered variables and the skewness increases. We propose the following work-flow for analyzing high-dimensional experiments with regions of altered variables: (1) Pre-process raw data using one of the standard normalization techniques. (2) Investigate if the distribution of the altered variables is skewed. (3) If the distribution is not believed to be skewed, no additional normalization is needed. Otherwise, re-normalize the data using a novel HMM-assisted normalization procedure. (4) Perform downstream analysis. Here, ChIP-chip data and simulated data were used to evaluate the performance of the work-flow. It was found that skewed distributions can be detected by using the novel DSE-test (Detection of Skewed Experiments). Furthermore, applying the HMM-assisted normalization to experiments where the distribution of the truly altered variables is skewed results in considerably higher sensitivity and lower bias than can be attained using standard and invariant normalization methods.

Classification of microarrays; synergistic effects between normalization, gene selection and machine learning.

BACKGROUND: Machine learning is a powerful approach for describing and predicting classes in microarray data. Although several comparative studies have investigated the relative performance of various machine learning methods, these often do not account for the fact that performance (e.g. error rate) is a result of a series of analysis steps of which the most important are data normalization, gene selection and machine learning. RESULTS: In this study, we used seven previously published cancer-related microarray data sets to compare the effects on classification performance of five normalization methods, three gene selection methods with 21 different numbers of selected genes and eight machine learning methods. Performance in term of error rate was rigorously estimated by repeatedly employing a double cross validation approach. Since performance varies greatly between data sets, we devised an analysis method that first compares methods within individual data sets and then visualizes the comparisons across data sets. We discovered both well performing individual methods and synergies between different methods. CONCLUSION: Support Vector Machines with a radial basis kernel, linear kernel or polynomial kernel of degree 2 all performed consistently well across data sets. We show that there is a synergistic relationship between these methods and gene selection based on the T-test and the selection of a relatively high number of genes. Also, we find that these methods benefit significantly from using normalized data, although it is hard to draw general conclusions about the relative performance of different normalization procedures.

Challenges in microarray class discovery: a comprehensive examination of normalization, gene selection and clustering.

BACKGROUND: Cluster analysis, and in particular hierarchical clustering, is widely used to extract information from gene expression data. The aim is to discover new classes, or sub-classes, of either individuals or genes. Performing a cluster analysis commonly involve decisions on how to; handle missing values, standardize the data and select genes. In addition, pre-processing, involving various types of filtration and normalization procedures, can have an effect on the ability to discover biologically relevant classes. Here we consider cluster analysis in a broad sense and perform a comprehensive evaluation that covers several aspects of cluster analyses, including normalization. RESULT: We evaluated 2780 cluster analysis methods on seven publicly available 2-channel microarray data sets with common reference designs. Each cluster analysis method differed in data normalization (5 normalizations were considered), missing value imputation (2), standardization of data (2), gene selection (19) or clustering method (11). The cluster analyses are evaluated using known classes, such as cancer types, and the adjusted Rand index. The performances of the different analyses vary between the data sets and it is difficult to give general recommendations. However, normalization, gene selection and clustering method are all variables that have a significant impact on the performance. In particular, gene selection is important and it is generally necessary to include a relatively large number of genes in order to get good performance. Selecting genes with high standard deviation or using principal component analysis are shown to be the preferred gene selection methods. Hierarchical clustering using Ward's method, k-means clustering and Mclust are the clustering methods considered in this paper that achieves the highest adjusted Rand. Normalization can have a significant positive impact on the ability to cluster individuals, and there are indications that background correction is preferable, in particular if the gene selection is successful. However, this is an area that needs to be studied further in order to draw any general conclusions. CONCLUSIONS: The choice of cluster analysis, and in particular gene selection, has a large impact on the ability to cluster individuals correctly based on expression profiles. Normalization has a positive effect, but the relative performance of different normalizations is an area that needs more research. In summary, although clustering, gene selection and normalization are considered standard methods in bioinformatics, our comprehensive analysis shows that selecting the right methods, and the right combinations of methods, is far from trivial and that much is still unexplored in what is considered to be the most basic analysis of genomic data.