Robert Nadon

Statistical issues with microarrays: processing and analysis

The study of gene expression with printed arrays and prefabricated chips is evolving from a qualitative to a quantitative science. Statistical procedures for determining quality control, differential expression, and reproducibility of findings are a natural consequence of this evolution. However, problems inherent to the technologies have raised important issues of how to apply adequate statistical tests. As a consequence, statistical approaches to microarray research are not yet as routine as they are in other sciences. Statistical methods, tailored to microarrays, continue to be adapted and developed. We present an overview of these methods and of outstanding issues in their use and validation.

Absorption and hypnotizability: context effects reexamined

Two independent studies failed to find evidence consistent with Council, Kirsch, and Hafner (1986), who argued that the repeatedly observed correlations between Tellegens (1981) Absorption Scale (TAS) and hypnosis measures were artifacts of testing context, and de Groot, Gwynn, and Spanos (1988), who claimed evidence for a Gender x Context moderator effect. In the present studies, Ss completed the TAS and other personality questionnaires on 2 occasions: during an independent survey and later immediately prior to an assessment of hypnotizability. In Experiment 1 (N = 475), the effect of context on the relation between questionnaire scores and hypnotizability was weak and variable; in Experiment 2 (N = 434), these weak effects were reversed. The results reaffirm the construct validity of absorption as both a major dimension of personality and as a predictor of hypnotic responsiveness.

Statistical inference methods for gene expression arrays

Gene expression arrays present unique challenges for statistical inference. They typically small number of replicated expression values in array studies make the use of standard parametric statistical tests problematic. Such test have low sensitivity and return potentially inaccurate probability values. This paper describes novel alternative statistical modeling procedures which circumvent these difficulties by pooling random error estimates obtained from replicate expression values. The procedures, which can be used with both micro- and macro-arrays, include outlier detection, confidence intervals, statistical test of differences between conditions, and statistical power analysis for determining number of replicates needed to detect between-condition differences of specified magnitude. The methods are illustrated with experimental data.

What this field needs is a good nomological network.

Research in the field of hypnosis lacks a coherent structure on which to build. This lack of a mature nomological network stems from fundamental disagreements concerning the construct validity of hypnotizability, which in turn stem in part from different research practices across laboratories. For these reasons, the field has had less impact on psychology and medicine than is warranted by the numerous sophisticated scientific studies that have been conducted during the past three decades.

Laterality of hypnotic response

In an investigation of hemispheric activity during hypnosis, a total of 1269 Ss received hypnotizability scales containing suggestions targeting the left or right side of the body. There were no consistent differences in response strength on the left compared to the right side. Nor were there differences in hypnotizability between right- and left-handed (and ambidextrous) Ss, or between Ss who sat on the left versus right side of the testing room. Definitive evidence of lateralized cerebral activity associated with hypnosis and hypnotizability can only come from direct neuropsychological, electrocortical, or brain-imaging investigations.

Estimation of a residual distribution with small numbers of repeated measurements

The authors consider the estimation of a residual distribution for different measurement problems with a common measurement error process. The problem is motivated by issues arising in the analysis of gene expression data but should have application in other similar settings. It is implicitly assumed through- out that there are large numbers of measurements but small numbers of repeated measurements. As a consequence, the distribution of the estimated residuals is a biased estimate of the residual distribution. The authors present two methods for the estimation of the residual distribution with some restriction on the form of the distribution. They give an upper bound for the rate of convergence for an estimator based on the characteristic function and compare its performance with that of another estimator with simulations.

Statistical informatics: software tools for statistical inference about expression arrays

65 cations to oncology research we have developed the AmpliOnc I array, which contains 3 spots for each of over 50 target loci (mostly oncogenes) that have been shown to be potentially amplified in tumour tissues. In hybridizations of normal female (green) against normal male (red) total human DNA, the spots containing the androgen receptor (X chromosome) revealed a green/red ratio significantly higher than the ratios of all other loci (>99% confidence level). This suggests sufficient sensitivity for single copy-change detection. In tests with defined tumour cell lines, the expected ratios were obtained for multiple amplifications with more than 95% confidence on each chip. Multiple tumour cell lines and primary tumours have been analysed with the AmpliOnc chip. In all cases known oncogene amplifications were confirmed, and in some instances additional and so far unknown amplifications were discovered.