GIANT working group home

Genomic Integration Across N Technologies

The GIANT working group facilitates interactions among people trained in statistics, bioinformatics, computer science, biostatistics, and medicine. Our collaborations are designed to integrate our various talents into a whole that is truly greater than the sum of its parts.

The new high-throughput technologies require a technical expertise unprecedented in the field. As we bring together more scientists with varied expertise, we are challenged to both use and create rigorous approaches to handling biological and clinical data.

SNPs, gene expression My research interests include statistical methods for combining high throughput gene expression data from multiple studies and characterizing chromosomal alterations in SNP arrays. In collaboration with statisticians at the Norwegian University of Science and Technology and the University of North Carolina, we have developed a Bayesian hierarchical model for estimating differential expression from multiple studies. For SNP arrays, we are developing hidden Markov models (HMMs) to characterize chromosomal alterations that may explain phenotypic variation in human diseases. Open source software developed for these methods include the R packages XDE (cross-study differential expression), SNPchip (visualization of genotype and copy number in high throughput SNP platforms) and VanillaICE (hidden Markov models for SNP chip platforms). Each of these packages are available from Bioconductor.	Figure legend: Predictions from the hidden Markov model {(N)ormal: two copies, (L)oss of heterozygosity, (A)mplification: > 2 copies) and (D)eletion: < 2 copies)} are plotted beneath the observed data displayed as log2 copy number versus physical position. The genotype calls are color coded by blue (AA or BB) and red (AB).
Cross-study Validation of Genomic Analyses One of our interests is measuring the reproducibility of gene expression patterns across studies. We use a tool called integrative correlation that is based on the cross study correlation of the within study gene-gene correlations (see left). When genes are scored for cross-study consistency of expression patterns using unsupervised integrative correlation, the reproducibility of subsequent supervised analyses is substantially enhanced. Above, integrative a hostogram if integrative correlations is shown at the top. At the bottom, data are split in high a and low integrative correlations. Each of the scatterplots at the bottom compare across studes the signal-to-noise ratios for each gene as a predictor of BRCA1 germline mutations. The subset of genes on the right (high integrative conrrelation) shows reproducibility while the other does not.
	The Transcriptome Landscape RNA is implicated in proteins biosynthesis as well as in several biological processes, which were not even imagined few years ago. RNA, indeed, is now recognized for its role in protein synthesis inhibition and for its contribution to genomic complexity: alternative transcription, splicing and translation and RNA editing allow single genes to encode for multiple proteins and a central role for non-coding RNA is beginning to emerge in cell signaling and regulation of gene expression. For these reasons my principal research interests and efforts are devoted to the understanding of Transcriptome complexity, by coupling DNA sequences annotation and gene expression profiling, supported by robust statistics and computation.
Unexpected tools: transposons Transposons are mobile genetic elements that are ubiquitous in life. Comprising various amounts of every genome sequenced to date, including nearly half of the human genome, their biology is mysteriously intertwined with ours. Transposons, especially retrotransposons and the L1 element, appear as culprits in many inherited diseases, but their presence is arguably less disruptive than expected, as their effects are dampened by many cellular systems. We have used them both as markers and as genetic tools, taking advantage of many of their unusual properties. Marked by their repeated sequences, they are somewhat easy to find but cause major snags in many high-throughput genomics projects.

GIANT Working Group

Home

Genomic Integration Across N Technologies