Stanford-South Africa Student Seminar Series

Wednesday, April 25, 2007, 9 AM PST, 6 PM CAT
CTL response to HIV-1 subtype C is poorly predicted by known epitope motifs

Nobubelo Ngandu, Cathal Seioghe

Cytotoxic T lymphocyte (CTL) responses are thought to be essential for the control of HIV-1 replication in vivo and immunogens that elicit CTL responses are currently a major focus of HIV vaccine research. Here we investigated two aspects of the CTL response to HIV-1 subtype C that are important for vaccine design and efficacy monitoring. First, we assessed the relationship between the CTL response and sequence diversity, using a robust statistical method. While peptides that were most frequently recognized by the CTL response in Nef and p24 tended to be conserved, this was not the case for p17 where epitope recognition coincided with highly variable regions. Secondly, we investigated the relationship between observed and predicted CTL responses, given the HLA genotype of infected individuals. Only 52% of the Nef peptides and 64% of the Gag peptides that elicited a CTL response contained sequence motifs thought to be required for binding by the HLA A or B alleles found in the corresponding patient. In a comparable subtype B dataset a much higher proportion of the peptides that elicited a CTL response were consistent with the patient HLA genotype (96% and 83% for Nef and Gag, respectively). We demonstrate that this difference between subtypes C and B is likely to result from a combination of a tendency for HLA alleles common in Southern African populations to be poorly characterized, as well as a tendency for sequence motifs associated with HLA recognition to be over-specified for sequence variation found in the B clade. Our results suggest that knowledge of HLA binding motifs is likely to be biased towards certain populations and subtypes. This can have important implications for understanding immune escape and predicting vaccine efficacy in the context of populations primarily infected with non-B subtype HIV-1.

Wednesday, April 11, 2007 9 AM PST, 6 PM CAT

Phylogenetic models of evolution: towards more accurate and useable
tools for characterising selection.

Konrad Scheffler, PhD, SSABMI Postdoctoral Visiting Scholar

Characterising the evolution of protein coding sequences is of wide
interest, providing clues for identifying important genes and
functional sites within genes. In the case of pathogens such as HIV,
detection of positively selected sites is of particular importance for
identifying potential drug targets and for understanding immune escape
and the emergence of drug resistance. For these reasons, there is an
increasing demand for easily useable tools that can analyse selective
pressure in sequence alignments. Despite this, existing analysis tools
are becoming ever more complicated due to theoretical advances which
allow simplifying assumptions to be relaxed and the resulting models
to become more realistic and accurate. This presentation will discuss
some of these advances and introduce PARRIS: A PARtitioning approach
for Robust Inference of Selection. With PARRIS we aim to make the
state of the art in phylogenetic modelling available without requiring
users to be experts in the field.

Wednesday 14 March 2007 at 19h00 CAT/ 09h00 Pacific:

MScanner: A tool to expand a corpus of Medline citations
Graham Poulter, Cathal Seoighe, Daniel Rubin, Russ Altman

Maintaining a comprehensive database of the literature in a given subject area is a difficult task to achieve manually, or even by reviewing keyword searches, because of the size and complexity of the biomedical literature. We have developed MScanner, a tool that uses a
probabilistic classifier to identify additional articles from Medline that are related to a collection of input articles. MScanner trains a naïve Bayes classifier on Medical Subject Heading (MeSH) term assignments and journal of publication, and assigns a score to every article in Medline. Articles scoring above a tunable threshold are classified as positive. Output includes a comprehensive set of performance statistics based on cross-validation as well as graphs showing the performance as a function of the score threshold and the score distributions of articles in the example set compared to the background. MScanner achieves areas under the ROC curves of approximately 0.98 to 0.99 using a variety of test data sets and can be used for literature annotation or to extend or update sets of related articles such as text mining corpora or collections of subject-specific articles.

Wednesday Feb. 21, 2007

HIV Drug Resistance Database

Robert Shafer, MD
Stanford University

Wednesday, June 7, 2006

Influence of gross evolutionary forces on the distribution and
expression of the Cancer / Testis genes in the human genome.

Sumir Panji, Winston Hide

The Cancer / Testis (CT) genes are a heterogeneous collation of human
genes whose principal unifying feature is an expression profile
pre-dominantly confined to the testis and a wide variety of cancer
types. The Cancer / Testis Antigens (CTAs) are a subset of the CT
genes that have been shown to elicit an immune response to neoplasms
that express them. The potential immunogenic properties of the CT
genes, their narrow tissue expression spectrum coupled with the
absence of any major histocomapatability complex expression in the
male germline cells have all contributed to the emergent attraction of
the CT genes as ideal cancer vaccination candidates. Due to the
attraction of the CT genes as cancer vaccines, characterisation of
their expression profiles and immunogenic properties dominate most of
the research conducted on the CT genes to date.

The human genome provides a coherent, functioning genomic framework
around which biological data can organized and systematically
analysed. The genomic locations of the CT genes were elucidated by the
mapping of CT gene transcripts to the human genome in order to
determine whether or not CT gene distribution in the human genome is
random. The distribution of the CT genes in the human genome was found
to be non-random with a bias towards to the X chromosome.

Current evolutionary models of the sex chromosomes predict that the X
chromosome should be depleted of testis expressed genes, contrary to
the enrichment of the CT genes observed on the X chromosome. A
competing evolutionary model, known as Rice’s hypothesis,
predicts the enrichment of testis specific genes on the X chromosome
under certain conditions. A parsimonious resolution between both
evolutionary theories that has been proposed based on the phase during
which X linked testis specific genes are expressed, is examined with
regards to CT gene expression.

Wednesday, May 24, 2006

Computational Modeling of Viral Protein Oligomers

Andres Bayani Tellez, Stanford BMI

The poliovirus replication complex involves the viral RNA-dependent RNA
polymerase in oligomeric complexes on the surface of membranes. The
building blocks of these oligomers are polymerase-polymerase interactions.
One polymerase-polymerase interaction site, Interface I, has been
identified and validated by previous work (Structure 5:1109; EMBO 20:1153,
JBC 277:31551). Interface I is an asymmetric interaction which forms
extendable head to tail fibers of polymerase. Interestingly, purified
polymerase forms large planar lattices, suggesting that a second oligomeric
interface exists on the polymerase which stacks up interface I fibers to
form the observed two-dimensional planar arrays. Identification of this
second interaction site remains an open question. To generate plausible
hypotheses about this second oligomeric interface, computational modeling
was employed.

The polymerase undergoes an allosteric change upon forming interface I (EMBO
23:3462). To model the alternative polymerase conformations sampled, low
frequency harmonic oscillations were calculated using normal mode analysis.
These conformations were modelled into polymerase-polymerase fibers, giving
rise to ten different conformations for a dimer along interface I. The
surface convolution was calculated for each pair of fibers to generate all
possible ways of combining the subunits, creating thousands of complexes.

The complexes were then clustered into groups according to the use of
particular residues. Of the resulting classes of tetramer, only the
parallel and anti-parallel sheets show potential for readily forming the
observed two dimensional lattices. Selection of the symmetric and
parsimonious interfaces within the family of parallel and anti-parallel
sheets resulted in four candidate interfaces involving distinct patches of
residues on the polymerase. Multiple sequence alignment of every known
picornavirus polymerase exhibits high conservation and co-variation within
the regions that make up the postulated second interface. Mutant polymerase
designed to disrupt the posited interactions will be purified and turbidity
assays will be performed for analysis of polymerization kinetics.

Wednesday May 3rd, 2006

Detection of positive selection in longitudinal HIV reverse
transcriptase sequences treated with nevirapine

Farahnaz Ketwaroo, Winston Hide, Cathal Seoighe, Konrad Scheffler

People living in the poorest countries of sub-Saharan Africa, where
HIV prevalence is also one of the highest in the world, do not have
access to expensive antiretroviral drugs to prevent mother to child
transmission (MTCT) of HIV. Nevirapine (NVP) is one of the few cost
effective drugs in the prevention of MTCT, and inhibits HIV enzyme
reverse transcriptase, even though transmission still occurs in 15-20%
of the cases. The transmission of HIV to newborns usually occurs at
the time of birth, which is also the time at which NVP is administered
[1]. HIV virions which pass on to newborns may have undergone positive
selection for viruses resistant to nevirapine.

Positive selection is inferred from sequences when the estimated rate
of nonsynonymous substitutions is significantly greater than the rate
of synonymous substitutions [2]. Commonly used methods for detecting
positive selection are not appropriate for sequences separated by time
intervals of a few months, for instance reverse transcriptase
sequences isolated prior to nevirapine treatment and sequences
persisting after treatment to NVP (in the mother and in the babies).

We implemented a novel approach for detecting positive selection in
reverse transcriptase sequences separated by short time points using
pairwise counting methods [3]. Three codon positions at which
mutations are known to confer strong NVP resistance (181, 103 and 106)
were found to be under positive selection in sequence pairs isolated
from 46 patients. In a larger dataset of 212 patients, a total of 8
sites, of which 4 are associated with NVP resistance, were found to be
under positive selection. Simulations were also performed to determine
the power and accuracy of the approach used. Based on these, the
approach implemented appears to be suitable to detect positive
selection in sequences separated by short time intervals.

References:
1. McIntyre J. Curr Opin Infect Dis. 2006 Feb;19(1):33-8.
2. Bielawski JP, Yang Z. Maximum likelihood methods for detecting
adaptive evolution after gene duplication. J Struct Funct Genomics.
2003;3(1-4):201-12.
3. Zhang J, Rosenberg HF, Nei M. Proc. Natl. Acad. Sci.
1998;95:3708-3713

Wednesday, April 19, 10:00AM

Chemogenomic Profiling Reveals Functional Networks in Yeast

Maureen E Hillenmeyer, Stanford University

The Yeast Deletion Collection contains ~6000 single-gene knockout strains, one for every gene in the genome. We have measured the fitness (growth) of these strains in ~300 drug conditions, using a high-throughput, quantitative assay called Chemogenomic Profiling. Each deletion strain has a resulting fitness profile across drug compounds, and each compound has an inhibition profile across strains; the data form a matrix of ~6000 genes (strains) by ~300 drugs. We have generated a “co-fitness” network of genes based on fitness profile similarity. Modules of related genes within the network demonstrate its utility in functional prediction. We are comparing this functional network to other networks, including co-expression, protein-protein interaction, and synthetic lethality. We perform a similar analysis of drugs across inhibition profiles, examining co-inhibiting drug groups for similarity in mechanisms of action. We hope to gain insight about drugs of unknown mechanism, and we plan to incorporate structural information in the future. Finally, we are integrating drug and gene information to make predictions about potential drug targets that can be experimentally validated.

References:
1. Shoemaker DD., et al. Nature Genetics: 14 (1996).
2. Winzeler EA, et al. Science: 285 (1999)
3. Giaever G, et al. Nature: 418 (2002)

Wednesday, April 5, 2006

Development and Implementation of a bridging ontology to allow for cross species gene expression analysis

Adele Kruger, University of the Western Cape

The direct comparison between gene expression events in mouse and human is not an entirely trivial task.  Current controlled vocabularies focus on the gene expression of their intended organisms, but to enable cross-species comparison, a standard vocabulary is required to enable direct mappings between the two species.  We have developed a mouse ontology to act as in intermediate between other available mouse ontologies and the human eVOC ontology.  This mouse ontology is based on the human ontology, and therefore allows the direct mapping between terms as required.  The ontologies are divided into 29 separate ontologies, one for each of the 28 Theiler stages of mouse development, and one for terms from unknown stages.  There are 2849 terms in the 29 ontologies, and they are cross-referenced to the EMAP and Mouse Anatomy ontologies, enabling interoperability between the ontologies.  These ontologies have already been used to investigate the expression of cancer testis genes during the development of mouse and human.  This investigation provided insight into the development of the genes themselves and the regulation of these genes during development.  The presentation will provide problems and solutions encountered with the development and implementation of the above mentioned intermediate ontology.

For further information about the seminar series, please contact:

Adele Kruger (South Africa)
adele <_AT_> sanbi.ac.za

or

Sarah Aerni
saerni <_AT_> stanford.edu

This work supported by the NIH/Fogarty International Center under grant D43 TW00699