Seminars at Rice meet from
2:00 pm to 3:00 pm in Duncan Hall 1075.
Exceptions
are noted in schedule below.
· 01 November
Professor Gorlov will discuss the following two papers.
1. Variation in the relative length of chromosomes in mammalian karyotypes. Hypothesis of equalizing selection. Gorlova OY, Gorlov IP. Genetika 2000 Jun;36(6):725-39
A hypothesis on the selective neutrality of relative lengths of karyotype chromosomes was tested. Idiograms expected based on an assumption of selective neutrality of chromosome lengths were compared with actual idiograms in more than a hundred mammalian species. The observed idiograms differed from those expected in a similar manner: in the observed idiograms, the longest chromosomes were shorter, and the shortest were longer than expected. It is suggested that karyotype chromosome variation is limited by selection against chromosome rearrangements that produce very long or very short chromosomes. Chromosome rearrangements tend to randomize relative chromosome lengths in a karyotype, whereas natural selection acts to equalize them.
There is well-known evidence that in many eukaryotes, different species have different karyotypes (e.g. n"1-47 in ants and n"3-51 in mammals). Alternative (fusion and fission) hypotheses have been proposed to interpret this chromosomal diversity. Although the former has long been accepted, accumulating molecular genetics evidence seems to support the latter. The authors investigated this problem from a stochastic viewpoint using the Monte Carlo simulation method under the minimum interaction theory. They found that the results of simulations consistently interpreted the chromosomal diversity observed in mammals, ants and wasps, and concluded that chromosome evolution tends to evolve as a whole toward increasing chromosome numbers by centric fission.
· 23 October, 3pm - 4pm Wednesday, Duncan Hall 1070
Note date, time and room change.
Using linkage disequilibrium (LD) to induce a covariance structure in a hierarchical model may help smooth out noise encountered in analysis of a complex disease. However, the common measures for LD can be noisy and may not produce a positive definite matrix necessary for a covariance measure. LD in the human leukocyte antigen (HLA) region of Chromosome 6 decays as a function of genetic distance, similar to covariance structures commonly encountered in spatial statistics. Therefore, we examine several Kriging Covariance Functions, commonly used in spatial statistics, as models for LD. Fitting the covaraince function to the LD measures between markers from the HLA region of Chromosome 6 gives us approximate LD measures between markers that also generate a positive definite matrix. Visual inspection of the predicted LD values shows that the rational quadratic semi-variogram captures many of the features of the real data, however, the fitted model also has an upward bias for LD between more distant markers, and we will discuss measures for goodness of fit of the model.Research supported, in part, by a Genetic Epidemiology Fellowship supported by NCI grant R25 CA57730, Dr. Robert Chamberlain, Ph.D. P.I.
·
18 OctoberMicroarray timecourse data present a difficult problem in unsupervised classification. Beyond the traditional estimation problems of model selection and choice of estimation criterion, the difficulties with microarray data are enriched by technical problems with the raw data. The talk will consider the relative merits of several clustering techniques with emphasis on the mixture model approach. Some discussion will be made of the scientific motivation for solving the clustering problem and some focus will be devoted to multi-timecourse data. Interaction between the speaker and the audience is to be encouraged.
· 04 October
I will give an introduction to serial analysis of gene expression (SAGE) technique and present some preliminary clustering analysis results on SAGE data. Before our work, hierarchical clustering method has been used to identify patterns in SAGE data and some interesting gene clusters were revealed by this method. We extend this idea to more clustering methods on more libraries in order to explore the performances of clustering methods in identifying gene clusters which might be co-regulated and contribute to the initialization and progression of certain type of cancer as well as evaluating the consistency of sample clustering with biological results. Five different clustering methods, namely, hierarchical clustering, K-means, Self-organizing map, principal component analysis and multidimensional scaling, are applied to 32 SAGE libraries with six different tissue types and library sizes varying from 20K to 100K. All analyses are carried out based on genes that are highly expressed across all libraries. The five methods give similar results for sample clustering. Some tissues distinguish themselves from others consistently in all methods while the classifications of a few tissue samples are ambiguous. Moreover, we find four clusters possibly associated with normal cells, colon cells, breast cells and leukocyte cell using hierarchical clustering, K-means and supervised-SOM. The first two approaches are more alike in finding gene clusters. Our work indicates that new methods need to be developed taking account of sampling error, size and quality of libraries, differences between data preparations from different labs and the effect of including more genes into analysis.
· 27 September
Authors: Andrzej Polanski, Adam Bobrowski and Marek Kimmel
We derive expressions of marginal distributions for times in the coalescence process with time - varying population size, which follow from the form of their joint probability distribution. Our method also allows computation of joint probability distribution for pairs, triples, etc. of coalescence times. The expressions derived are useful for extending several statistics from time constant to time varying case, increasing efficiency and accuracy of simulations in time varying evolution and debugging coalescence simulation software. For large genealogies (of size n > 50) the described method cannot be directly applied because expressions contain coefficients which become very large. This effect considerably limits practical applications of the described approach. Nevertheless we can show that some important statistics can be computed by using techniques of hypergeometric series summation.
· 20 September
I will discuss a new paper by Stephan Wooding and Alan Rogers. The authors propose a method for using genome-wide SNP data to test hypothesis of population size history. The method employs coalescent theory to derive the SNP frequency spectrum implied by a particular hypothesis, and then uses this frequency spectrum to assign a likelihood to a given set of data. I will contrast this method with existing methods for population size inference from sequence data.
Stephan Wooding and Alan Rogers (2002). The Matrix Coalescent
and an Application to Human Single-Nucleotide Polymorphisms. Genetics
161: 1641-1650.
· 13 September
I will present new adaptations on Gibbs Sampler and Metropolis-Hasting Algorithm with concern of reduction of asymptotic risk. Markov Chain Monte Carlo (MCMC) routines have become a fundamental means for generating random variants from distributions which are difficult to sample. The Hastings sampler, which includes the Gibbs and Metropolis samplers as special cases, is the most popular MCMC method. A number of adaptations are available to perform a variety of analyses and inference task. And many adaptations have been made to improve both the efficiency and accuracy of samplers from MCMC. In this talk I will present work in this area using both simulated data and real data.