| |||||||||||||||
 | |||||||||||||||
Interview SeminarsBioinformatics and Computational Biology Faculty Position
DAVID R. BICKEL Title:
Abstract: Recent advances in empirical Bayes methodologycit_bf1-cit_af ref_bf(Bickel, 2004 (-) ref_num88)ref_afcit_bf2cit_af ref_bf(Bickel, 2004 ref_num69)ref_af and network inferencecit_bf3cit_af ref_bf(Bickel, 2005 ref_num71)ref_af have increased the reliability of biological interpretations of gene expression measurements. Inference in computational biology may be further improved by research on statistically integrating various sources of information in addition to gene expression. The rationale behind statistically integrating information from different sources is that though each source of data in itself may provide little evidence for or against a hypothesis of interest, the simultaneous consideration of data from multiple sources can often enable the researcher to reasonably draw a conclusion with some quantifiably high degree of confidence. In the case of gene expression studies, since the data are highly variable and the number of biological replicates is usually small, information outside the expression measurements, such as expression measurements from previous experiments and the level of confidence that a gene of interest has a given function, can lead to substantially improved conclusions, for example, about an underlying biomolecular network of interactions. (The degree of uncertainty in those conclusions may be quantified consistently by the modern computational methods used in applied Bayesian statistics, which, unlike classical hypothesis testing, can quantify evidence in favor of a point hypothesis as naturally as evidence against it.) Unfortunately, such additional information is typically only used to validate the results of statistical tests or sophisticated computational methods, if at all. By contrast, recent validation algorithms from the machine learning community do not require complete separation of a test data set from a training data set and thus can complement the featured Bayesian approaches to solving complex problems of data integration and network reconstruction. (2) Bickel, D. R. (2004) “Degrees of differential gene expression: Detecting biologically significant expression differences and estimating their magnitudes,” Bioinformatics 20, 682-688.
IDDO FRIEDBERG Title: Abstract: Proteins mediate the lion's share of the functions necessary to maintain life. The diversity of functions proteins can perform is truly amazing and it is by studying protein structures that we discover how these functions are facilitated. However the computational study of protein structures can also reveal evolutionary and biophysical findings. My talk will be concerned with several aspects of the structure and function of proteins.Time permitting, I will also discuss a new strategy for target selection for structural genomics. The Protein Structure Initiative (PSI) is a broad initiative of various centers aiming to provide a complete coverage of protein structure space. Since it is not feasible to experimentally determine the structures of all proteins, it is generally agreed that the only viable strategy to achieve such coverage is to carefully select specific proteins (“targets”), determine their structure experimentally, and then use comparative modeling techniques to model the rest. We suggest that the structural genomics community, in addition to any adopted target selection strategy, should also take care to study representatives of families that are predicted to have significant functional variations within known structural fold groups. The reason being, that a structural template may be falsely taken to be a functional template as well for modeling. We have developed a method for doing so, and I will discuss some of our findings.
MAXIM BAZHENOV Title: Abstract: The olfactory system maps complex and high dimensional olfactory stimuli (odors) into unique and reproducible ensembles of neuronal activity. This mapping includes multilevel processing and involves complex strategies for the efficient encoding of information. In the olfactory system of insects, dense and dynamic odor representations generated in the antennal lobe (functional analog of the olfactory bulb) are transformed into sparse and specific patterns of neuronal activity in the mushroom body (analog of olfactory cortex). Remarkably, odor representations in the mushroom body remain sparse over thousand-fold changes in odor concentration, a feature potentially useful for storing and retrieving memories. Drawing on results obtained with biophysical network models of the olfactory system, I will discuss intrinsic and circuit properties that contribute to encoding olfactory information at different levels of odor processing, and the role of the intrinsic dynamics of the olfactory system in optimizing odor representations. I will also present a novel hypothesis that may reveal how competition between excitation and inhibition in the olfactory system creates a gain control mechanism for maintaining the stability and sparseness of neural codes for odors across broad ranges of concentrations.
CHIA-EN ANGELINA CHANG Title: Abstract: Computer modeling is used for understanding protein function and for providing details from the atomistic level to large-scale conformational changes. I will first introduce multiscale modeling methods (coarse-grained Brownian dynamics simulations and all-atom molecular dynamics simulations in implicit solvent), and their applications to biological systems. The HIV-1 protease internal motions, as well as the binding processes of ligands as they enter the binding site of the HIV-1 protease, will be discussed. A recent method of computing binding free energies will then be presented. This method uses detailed all-atom models. The validation studies on small model systems, and unexpected insights regarding the changes in configurational entropy upon binding, will be discussed. JUN-TAO GUO Title: Abstract: Knowledge of the 3-D structure of a protein is essential to understanding its function(s). Due to the long and expensive processes required for experimental protein structure determination, the disparities between the number of solved protein structures and the number of known protein sequences continue to grow. Computational structure prediction and modeling represents a valuable and feasible approach to narrowing the gap and to maximizing the biological information encrypted in the amino acid sequences. In this talk, I will first introduce the development of our protein structure prediction program, PROSPECT, which employs a threading methodology to solve the sequence-structure alignment problem. The biological application of protein modeling in three important research areas will then be discussed: (1) modeling of the core of amyloid fibril structures; (2) structure prediction of alternatively spliced protein isoforms; and (3) DNA binding motif prediction by protein-DNA docking techniques and its potential in structure-based systems biology.
XIAOHUI XIE Title: Abstract: Among the 3 billion bases contained in the human genome, only 1.5% are well characterized, primarily in the form of protein-coding genes. One of the main challenges in genomics is to understand the function of the other 98.5% of the genome. Comparison of the human genome to several other related genomes has revealed that these regions harbor a strikingly large number of highly conserved noncoding elements, accounting for over two-thirds of the portion of the human genome under selection. And yet the function of these conserved noncoding elements (CNEs) remains largely unknown. We also know little about their evolutionary origins, or the molecular mechanisms that have preserved them through millions of years’ evolution. I will describe computational methods for systematically dissecting the function of the CNEs. Using statistical analysis and comparative genomics, we have uncovered hundreds of novel regulatory motifs within the CNEs, matching hundreds of thousands of conserved instances in the genome. These motifs form distinct classes, including transcriptional regulatory elements, small RNA genes, microRNA targeting sites, and chromatin barriers. I will also describe an effort to characterize the evolution of regulatory sequences. I will propose the creative role of transposable elements as a major force for duplicating and dispersing regulatory elements in the human genome. Comparison of metatherian and eutherian genomes reveals that over 15% of the eutherian CNEs arose from sequence inserted by transposons. In a few years, genome sequences of over 50 mammals will become available. I will discuss how these data will empower the methods I have described, and provide us an opportunity to unravel all information coded in the human genome.
JOSHUA REST Title: Abstract: Gene expression is controlled by transcription factors that bind to a variable family of motifs in a gene’s promoter; however, the role of variability within motif families has not been systematically studied. Deciphering the function of these variable positions is important for understanding the complex cis-regulatory code that underlies the physiology, ecology and evolution of gene expression. Using a computational approach, I show that functional variants of these motifs among genes within the yeast genome are associated with condition-specific differential expression. The regulatory consequences of these variants are often conserved across different yeast species, and their evolution is associated with change in gene expression. The power of binding site variants to predict gene expression is tested by experimentally switching between binding site variants through site-directed mutagenesis in yeast.
HAIXU TANG Title: Abstract: Shotgun proteomics refers to the use of bottom-up proteomics techniques in which the protein content in a biological sample mixture is digested followed by high throughput liquid chromatography mass spectrometry (LC/MS) analysis. The resulting peptide fragment spectra are searched against a protein database to identify peptides in the sample using a computer program such as Sequest or Mascot. Finally, the identified peptides will be assembled into proteins that these peptides come from. In this talk, I will introduce several computational methods to improve this common data analysis protocol. First, we developed a new approach to peptide identification using the predicted theoretical peptide fragmentation in mass spectrometer. Next, we revisited the problem of inferring proteins in the sample from the peptide identification results and proposed a solution to this problem based on a novel notion called peptide detectability. Finally, we applied the idea of peptide detectability to the problem of protein quantification, i.e. to determine the abundances of proteins in the sample.
|
CEPCEB Home
rtz media
maintained by:
webmaster
last modified: