Technometrics Highlights
August Issue Covers Quality Assessment
for Microarrays
T
he advent of microarray technology Rank Approximation of Data Matrices
in the last 15 years has opened up with Element-wise Contamination.” Their
completely new avenues of research method gives a low-rank approximation
in the biological and medical sciences. that is resistant against the existence of
Where earlier research might have focused both atypical rows and scattered atypical
on a small number of carefully selected cells and is able to cope with missing val-
genes, today’s scientists simultaneously ues. Their algorithm is based on alternat-
study the importance of thousands of ing M-regressions and a starting estimate
genes, each corresponding to a spot on a from successive rank-one fits. They find
microarray chip. The vast quantities of data the method is effective both in simulations
provided by a microarray pose a number of and on actual data sets.
challenging statistical problems. Our lead Yi Fang and Myong K. Jeong consider
article—by Julia Brettschneider, François the problem of identifying linear relation-
Collin, Benjamin M. Bolstad, and ships between two sets of observed vari-
Terence P. Speed—highlights the crucial ables contaminated with outliers in their
area of quality assessment of the experi- paper, “Robust Probabilistic Multivariate
mental data. Their article presents methods Calibration Model.” Their RPMC approach
for visualizing outliers and trends and for uses the multivariate Student’s t distribu-
identifying poor-quality arrays or variable- tion, rather than the Gaussian, for noises
quality sets of arrays. They introduce sev- and latent variables. The authors derive an
eral quality measures based on probe-level efficient EM algorithm for parameter esti-
and probeset-level information, all obtained mation in RPMC and a procedure for the
number of criteria can be used for blind
as a byproduct of standard, low-level analy- detection of outliers. Experimental results
separation, and a method is developed here
sis algorithms. The article relates ideas from both simulated examples and real-life
for automatically combining them. The
developed in the context of quality control data sets show the effectiveness and robust-
resulting algorithm is more effective and
by W. A. Shewhart to this exciting branch ness of the proposed approach.
robust than counterparts that use other
of biological research. A number of actual The spectral density of a process is
means of combining the criteria.
data sets are used to illustrate the ideas. often used to characterize its dynamic
Quality control charts often begin
The paper is accompanied by five discus- behavior and can be a useful tool for com-
with a phase I sample of data that is pre-
sions and a rejoinder from the authors. paring the behavior of several processes.
sumed to be in statistical control and can
Multivariate data forms a theme running Konstantinos Fokianos and Alexios
be used to assess stable process character-
through many of the papers in this issue. Savvides address this problem in their
istics. Thus, it is important to detect signs
Three papers explicitly develop methods paper, “On Comparing Several Spectral
of unstable behavior in phase I data. This
for multivariate problems, and several oth- Densities.” They propose a novel semi-
problem is heightened when the process
ers are motivated by problems associated parametric loglinear model that links all
has several stages, with data collected at
with large and high-dimensional data. the spectral densities under consideration.
each one. Changliang Zou, Fugee Tsung,
Irene Epifanio considers the problem of This representation provides a parametric
and Yukun Liu address this problem in
curve discrimination in her paper, “Shape form for comparing spectral densities and
“A Change Point Approach for Phase I
Descriptors for Classification of Functional thus permits use of large-sample theory for
Analysis in Multistage Processes.” They
Data.” She proposes several shape descrip- maximum likelihood estimators as a basis
argue that particular types of changes are
tors for classifying functional data, includ- for statistical inference.
more likely to occur in multistage process-
ing use of statistical moments, coefficients Manufacturing processes often gener-
es. They use that assumption to develop
from independent component analysis, and ate large amounts of multivariate measure-
methods that are more sensitive to detect-
two mathematical morphology descriptors ment and inspection data. Xumei Shan
ing the most likely changes.
(morphological covariance and spatial size and Daniel W. Apley develop methods for
The Generalized Likelihood Ratio
distributions). She also proposes a meth- using these data to track down and elimi-
(GLR) test is another method for change
od inspired by work on form analysis in nate root causes of manufacturing varia-
point detection in process monitoring and
image processing. tion in their article, “Blind Identification
is especially effective when the data are
Ricardo A. Maronna and Víctor J. of Manufacturing Variation Patterns by
autocorrelated. Giovanna Capizzi and
Yohai present a method for “Robust Low Combining Source Separation Criteria.” A
Guido Masarotto address the problem of
AUGUST 2008 AMSTAT NEWS 13
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42 |
Page 43 |
Page 44 |
Page 45 |
Page 46 |
Page 47 |
Page 48 |
Page 49 |
Page 50 |
Page 51 |
Page 52 |
Page 53 |
Page 54 |
Page 55 |
Page 56 |
Page 57 |
Page 58 |
Page 59 |
Page 60 |
Page 61 |
Page 62 |
Page 63 |
Page 64 |
Page 65 |
Page 66 |
Page 67 |
Page 68