RECOMB Satellite Conferences on Systems Biology and Computational Proteomics was also this weekend here at UCSD. Although I will not outline all the talks, they were very good and had a number of very prominent and interesting speakers.
Daphne Koller, Stanford. Genetic Variation and Regulatory Networks: Mechanisms and Complexity. Her lab was looking at Yeast variation and gene expression data from Kruglyak Lab tool called Geronemo (Genetic RegulatOry Network of Modules). Basically an extension of their previous Bayesian network algorithm to handle genotypes. Using this approach they can find genotype-expression relationships that wouldnt be found by traditional association tests. She presented examples of modules that contain a large number of sequential of telomeric genes. Also a Puf3 module, containing gcn20 which seems to act together with puf3. Using diverse genomic data they can connect multiple members of this module together
Bing Ren, Decoding the human genome, a chip-chip approach. 13,804 CDCF insulator binding sites mapped. These elements block enhancer affect on promoters that it lies between. These sites coincide with boundaries of gene clusters that escaped X-inactivation. Also find enhancers that bind over 2000 human TFs, they look at co-activators of these elements including P300, CBP, TRAP220, PCAF, BRG-1, BAF170, TIP60, P160 in human ENCODE regions. Using patterns of histone modifications made a predictor of promoters or enhancers. This reveals a number of new predicted elements. They are validating a number of these experimentally. Whether this code is conserved is still a matter of inquiry.
Ron Shamir, Tel-Aviv University. Modeling and Expansion of Signaling Pathways. He said 5 years ago he was very naïve and thought with microarays could reconstruct networks de novo. Now theres 100x more data but still hopeless to reconstruct networks de novo. In this work they use a three state model to identify relations between genes using knockout gene expression measurements.
Mark Gerstein, Yale, Understanding protein function of a genome-scale using Networks. Used co-crystal structures of interaction partners and split then into mutually exclusive or simultaneous depending on if the interfaces overlap. This process gives about 1200 interactions and supports findings of date hubs and party hubs. These hubs with multiple interfaces are more likely to be essential, and have highly correlated expression among their partners, and have lower evolutionary rate. Hubs tend to gain interaction by duplication of their partners, this would best fit where the protein with a single interface is re-used, rather than a whole new interface.
Hierarchy of TFs in yeast depending on the genes and other TFs they regulate. The TFs at the lower part of the hierarchy are most essential (they are doing the work basically). Using knockout gene expression data, the higher level TFs had the most effect on the number genes changed. Also TFs at the top of the regulatory hierarchy have the highest degree in PPI networks. There was some tense moments during Q&A where methods of the hierarchy construction were scrutinized. Seems to me that their results are very sensitive to heirarchy construction and that in looped networks these heirarchys cannot be determined.
In network motifs found that co-regulated genes tend to interact, and they are more likely to interact the higher number of shared regulators. One exception if that there is long-range co-regulation in metabolic networks, basically that interacting regulators regulate metabolic enzymes that are not directly interacting.
David Eisenberg, UCLA. Cultural learning of computational biology for make benefit glorious science structural. BEST TITLE EVER! Find protein complex crystallization failures, use the string database to predict their complex partners and purify both together for crystallization. Structure of amyloid fibrils, altered a prion protein Sub35, and it only requires 6-4 residues to form he fibirils. Found other amyloid forming sequences based on computational design (rosetta) found that sequences need self-complementarity to form the “steric zipper” of fibrils.
Steven Briggs, UCSD. Large-scale proteomics from plants to humans. Used MS to identify 18,300 proteins using traditional methods, InSpect finds more 40% more spectra and 70% more peptides. Use spectral counts to quantify protein levels between people with and without CLL. 230 were upregulated and 71 downregulated in aggressive CLL patients. To tease out signaling pathways identify phospho-proteins and then work backword to reconstruct signaling pathways.
Nevan Krogan, UCSF. Biology without bias: functional insights from high resolution genetic and protein-protein interaction maps. He first presented a new analysis of the yeast TAP-tag complexes data that integrates both studies (Krogan and Gavin) using a bayesian purification element score that takes into account bait-prey and prey-prey interactions. Complexes using this method are highly co-localizaed, co-expressed and co-functional. He uses heirarchical clustering to make pictures of protein complexes, and apparently is a big fan of this method of visualization. He also presented preliminary E-MAP data which is a way to screen double deletions for sensitivity of redundancy (epistasis). This includes a 750x750 cross for chromosomal function. In this study even essential genes can be modulated but not deleted by DAMP mRNA repressor to understand their function. The data indicate that patterns of genetic interactions and protein interactions work together between them (epistatic genetic interactions) show proteins in the same pathway and are more likely to interact. He is also working on various other maps.
In all it was a good conference, with great weather and good talks. I was unable to go the last day about "Computational Proteomics", but I welcome anyone to leave comments about that. The auditorium was about full the whole time so others seemed interested as well. I urge anyone interested in this field to attend next year!