Arnoldo Frigessi

Predicting survival from microarray data—a comparative study

MOTIVATION Survival prediction from gene expression data and other high-dimensional genomic data has been subject to much research during the last years. These kinds of data are associated with the methodological problem of having many more gene expression values than individuals. In addition, the responses are censored survival times. Most of the proposed methods handle this by using Coxs proportional hazards model and obtain parameter estimates by some dimension reduction or parameter shrinkage estimation technique. Using three well-known microarray gene expression data sets, we compare the prediction performance of seven such methods: univariate selection, forward stepwise selection, principal components regression (PCR), supervised principal components regression, partial least squares regression (PLS), ridge regression and the lasso. RESULTS Statistical learning from subsets should be repeated several times in order to get a fair comparison between methods. Methods using coefficient shrinkage or linear combinations of the gene expression values have much better performance than the simple variable selection methods. For our data sets, ridge regression has the overall best performance. AVAILABILITY Matlab and R code for the prediction methods are available at http://www.med.uio.no/imb/stat/bmms/software/microsurv/.

Principles and methods of integrative genomic analyses in cancer

Combined analyses of molecular data, such as DNA copy-number alteration, mRNA and protein expression, point to biological functions and molecular pathways being deregulated in multiple cancers. Genomic, metabolomic and clinical data from various solid cancers and model systems are emerging and can be used to identify novel patient subgroups for tailored therapy and monitoring. The integrative genomics methodologies that are used to interpret these data require expertise in different disciplines, such as biology, medicine, mathematics, statistics and bioinformatics, and they can seem daunting. The objectives, methods and computational tools of integrative genomics that are available to date are reviewed here, as is their implementation in cancer research.

Regional thinning of the cerebral cortex in schizophrenia : Effects of diagnosis, age and antipsychotic medication

Morphological abnormalities of the cerebral cortex have been reported in a number of MRI-studies in schizophrenia. Uncertainty remains regarding cause, mechanism and progression of the alterations. It has been suggested that antipsychotic medication reduces total gray matter volumes, but results are inconsistent. In the present study differences in regional cortical thickness between 96 patients with a DSM-IV diagnosis of schizophrenia (n=81) or schizoaffective disorder (n=15) and 107 healthy subjects (mean age 42 years, range 17-57 years) were investigated using MRI and computer image analysis. Cortical thickness was estimated as the shortest distance between the gray/white matter border and the pial surface at numerous points across the entire cortical mantle. The influence of age and antipsychotic medication on variation in global and regional cortical thickness was explored. Thinner cortex among patients than controls was found in prefrontal and temporal regions of both hemispheres, while parietal and occipital regions were relatively spared. Some hemispheric specificity was noted, as regions of the prefrontal cortex were more affected in the right hemisphere, and regions of the temporal cortex in the left hemisphere. No significant interaction effect of age and diagnostic group on variation in cortical thickness was demonstrated. Among patients, dose or type of antipsychotic medication did not affect variation in cortical thickness. The results from this hitherto largest study on the topic show that prefrontal and temporal cortical thinning in patients with schizophrenia compared to controls is as pronounced in older as in younger subjects. The lack of significant influence from antipsychotic medication supports that regional cortical thinning is an inherent feature of the neurobiological disease process in schizophrenia.

On the simplified pair-copula construction - Simply useful or too simplistic?

Due to their high flexibility, yet simple structure, pair-copula constructions (PCCs) are becoming increasingly popular for constructing continuous multivariate distributions. However, inference requires the simplifying assumption that all the pair-copulae depend on the conditioning variables merely through the two conditional distribution functions that constitute their arguments, and not directly. In terms of standard measures of dependence, we express conditions under which a specific pair-copula decomposition of a multivariate distribution is of this simplified form. Moreover, we show that the simplified PCC in fact is a rather good approximation, even when the simplifying assumption is far from being fulfilled by the actual model.

Brain-derived neurotrophic factor gene (BDNF) variants and schizophrenia: An association study

Polymorphisms in the brain-derived neurotrophic factor (BDNF) gene have been suggested to be associated with schizophrenia. In a replication attempt, Swedish patients with schizophrenia (n=187) and control subjects (n=275) were assessed for four BDNF gene polymorphisms. There were no significantly different allele, genotype or haplotype frequencies between cases or controls. Neither were there any differences when schizophrenic patients were sub-divided with regard to a number of different clinical variables, although a small group of psychotic patients with prominent affective features displayed higher frequencies of the less common alleles of the Val66Met and 11757 G/C polymorphisms compared to controls. The present Swedish results do not verify previous associations between putative functional BDNF gene polymorphisms and schizophrenia. However, when combined with previous studies meta-analyses indicated that the BDNF 270 T-allele and the Val66Met homozygous state were associated with the disorder. Thus, the BDNF gene may confer susceptibility to schizophrenia. Additional studies are warranted to shed further light on this possibility.

The Genomic HyperBrowser: inferential genomics at the sequence level

The immense increase in the generation of genomic scale data poses an unmet analytical challenge, due to a lack of established methodology with the required flexibility and power. We propose a first principled approach to statistical analysis of sequence-level genomic information. We provide a growing collection of generic biological investigations that query pairwise relations between tracks, represented as mathematical objects, along the genome. The Genomic HyperBrowser implements the approach and is available at http://hyperbrowser.uio.no.

Genome-wide DNA methylation profiles in progression to in situ and invasive carcinoma of the breast with impact on gene transcription and prognosis

BackgroundDuctal carcinoma in situ (DCIS) of the breast is a precursor of invasive breast carcinoma. DNA methylation alterations are thought to be an early event in progression of cancer, and may prove valuable as a tool in clinical decision making and for understanding neoplastic development.ResultsWe generate genome-wide DNA methylation profiles of 285 breast tissue samples representing progression of cancer, and validate methylation changes between normal and DCIS in an independent dataset of 15 normal and 40 DCIS samples. We also validate a prognostic signature on 583 breast cancer samples from The Cancer Genome Atlas. Our analysis reveals that DNA methylation profiles of DCIS are radically altered compared to normal breast tissue, involving more than 5,000 genes. Changes between DCIS and invasive breast carcinoma involve around 1,000 genes. In tumors, DNA methylation is associated with gene expression of almost 3,000 genes, including both negative and positive correlations. A prognostic signature based on methylation level of 18 CpGs is associated with survival of breast cancer patients with invasive tumors, as well as with survival of patients with DCIS and mixed lesions of DCIS and invasive breast carcinoma.ConclusionsThis work demonstrates that changes in the epigenome occur early in the neoplastic progression, provides evidence for the possible utilization of DNA methylation-based markers of progression in the clinic, and highlights the importance of epigenetic changes in carcinogenesis.

BDNF gene variants and brain morphology in schizophrenia.

The objective was to investigate putative associations between brain‐derived neurotrophic factor gene (BDNF) polymorphisms and brain morphology in patients with schizophrenia and healthy control subjects. Four BDNF polymorphisms were genotyped and analyzed versus 39 brain volume measures in 96 patients with schizophrenia or schizoaffective disorder and 104 healthy subjects. In all subjects, quantitative data on segmented gray, white, and cerebrospinal fluid (CSF) tissue class volumes of total brain and major cerebral lobes including ventricular CSF were obtained using magnetic resonance imaging (MRI). In a randomly selected subset of this population (n = 101–122), MR volumes from cerebellar tonsil, hemispheres, and vermis subregions, striatal structures, hippocampus, and corpus callosum were also measured. The BDNF 11757 G/C polymorphism was highly significantly associated with frontal gray matter volume variation in patients alone and in patients and control subjects combined. In patients only, the 270 C/T polymorphism was associated with total caudate volume. Significant associations were demonstrated between the BDNF 11757 G/C and Val66Met polymorphisms and a global haplotype estimate of four BDNF polymorphisms and the posterior superior cerebellar vermis volume in the controls as well as in the combined group, but not in the patients. The 11757 G/C polymorphism was associated with cerebellar hemisphere white and gray matter volumes in the combined group. The BDNF −633 T/A polymorphism was associated with gray matter of the putamen in the controls. Trends for associations between several polymorphisms/haplotype estimates and MRI volumes were found. BDNF gene variation may influence brain morphology. The effects may be different in patients with schizophrenia and healthy subjects.

Climatically driven synchrony of gerbil populations allows large-scale plague outbreaks

In central Asia, the great gerbil (Rhombomys opimus) is the main host for the bacterium Yersinia pestis, the cause of bubonic plague. In order to prevent plague outbreaks, monitoring of the great gerbil has been carried out in Kazakhstan since the late 1940s. We use the resulting data to demonstrate that climate forcing synchronizes the dynamics of gerbils over large geographical areas. As it is known that gerbil densities need to exceed a threshold level for plague to persist, synchrony in gerbil abundance across large geographical areas is likely to be a condition for plague outbreaks at similar large scales. Here, we substantiate this proposition through autoregressive modelling involving the normalized differentiated vegetation index as a forcing covariate. Based upon predicted climate changes, our study suggests that during the next century, plague epizootics may become more frequent in central Asia.

Eight genes are highly associated with BMD variation in postmenopausal Caucasian women

Low bone mineral density (BMD) is an important risk factor for skeletal fractures which occur in about 40% of women >/=50 years in the western world. We describe the transcriptional changes in 84 trans-iliacal bone biopsies associated with BMD variations in postmenopausal females (50 to 86 years), aiming to identify genetic determinants of bone structure. The women were healthy or having a primary osteopenic or osteoporotic status with or without low energy fractures. The total cohort of 91 unrelated women representing a wide range of BMDs, were consecutively registered and submitted to global gene Affymetrix microarray expression analysis or histomorphometry. Among almost 23,000 expressed transcripts, a set represented by ACSL3 (acyl-CoA synthetase long-chain family member 3), NIPSNAP3B (nipsnap homolog 3B), DLEU2 (Deleted in lymphocytic leukemia, 2), C1ORF61 (Chromosome 1 open reading frame 61), DKK1 (Dickkopf homolog 1), SOST (Sclerostin), ABCA8, (ATP-binding cassette, sub-family A, member 8), and uncharacterized (AFFX-M27830-M-at), was significantly correlated to total hip BMD (5% false discovery rate) explaining 62% of the BMD variation expressed as T-score, 53% when adjusting for the influence of age (Z-score) and 44% when further adjusting for body mass index (BMI). Only SOST was previously associated to BMD, and the majority of the genes have previously not been associated with a bone phenotype. In molecular network analyses, SOST shows a strong, positive correlation with DKK1, both being members of the Wnt signaling pathway. The results provide novel insight in the underlying biology of bone metabolism and osteoporosis which is the ultimate consequence of low BMD.