|
|
|
|
| |
|
[ Jump to Conferences ] SBC seminar series, 2013Every second Tuesday, at 10:30.
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| Tuesday June 4 | 10:30 | Satish Nair | Stockholm University |
| Lunch room at Scilifelab, floor 2 | Master thesis presentation | ||
| Abstract soon to come. | |||
| Tuesday June 4 | ~11:00 | Daniele Raimondi | Stockholm University |
| Lunch room at Scilifelab, floor 2 | Master thesis presentation: Deep learning ensemble methodology for direct information contact prediction | ||
|
Recently, several new contact prediction methods have been published.
They use (i) large sets of multiple aligned sequences (ii) and assume that
correlations between columns in these alignments can be the results of
residue interactions and thus clues of residues spatial proximity in the
native structure. These methods are clearly superior to earlier methods
when it comes to predicting contacts in proteins. PconsC [2] has been
developed by Marcin J. Skwark and combines predictions from two direct
information methods, PSICOV [4] and plmDCA [3], and two alignment
methods, HHblits and jackHmmer, at four different e-value thresholds,
obtaining an improvement of the predictive performances with respect to
the single methods on which it is based. The aim of this thesis project
was to further improve the quality of these predictions. To achieve this goal, I developed a Deep Learning architecture able of performing structured predictions, taking into consideration the significant amount of information underlying the contact prediction problem instead of simply considering each residue pair independent from the others (in [1] has been shown how contacts in the native structure can hardly involve a single pair of residues). I implemented a multilayer learner using Random Forest classiffers that improves contact predictions by being able to abstract some typical inter/multi residue relationships among neighbouring residue pairs, namely by learning how to recognize frequent visual patterns (mainly Secondary Structure features, such as alfa-helices and beta-sheets) in the contact maps. This abstraction ability can relocate the most uncertain predictions into the recognized patterns, reconstructing them and thus improving significantly the precision of the overall contact map. This Deep Learning approach, along with some additional features (e.g. predicted Secondary Structure and predicted Relative Solvent Accessibility) can provide a further 20% improvement of PconsC predictive performances. References [1] Pietro Di Lena, Ken Nagata and Pierre Baldi, Deep architectures for protein contact map prediction Vol. 28 no. 19 2012, pages 2449 2457 BIOINFOR- MATICS doi:10.1093/bioinformatics/bts475 [2] Marcin J. Skwark, Abbi Abdel-Rehim and Arne Elofsson, PconsC: Combi- nation of direct information methods and alignments improves contact pre- diction, Bioinformatics (2013) doi: 10.1093/bioinformatics/btt259 [3] Ekeberg, M., Lovkvist, C., Lan, Y., Weigt, M., and Aurell, E. (2013). Im- proved contact prediction in proteins: Using pseudolikelihoods to infer Potts models, Phys Rev E Stat Nonlin Soft Matter Phys, 87(1-1), 012707. [4] Jones, D., Buchan, D., Cozzetto, D., and Pontil, M. (2012), PSICOV: pre- cise structural contact prediction using sparse inverse covariance estimation on large multiple sequence alignments, Bioinformatics, 28(2), 184 190. | |||
| Tuesday June 11 | 10:30 | Teepo Niinimäki | University of Helsinki, Helsinki, Finland |
| Lunch room at Scilifelab, floor 2 | |||
| Abstract soon to come. | |||
| Tuesday September 3 | 10:30 | Michael Y. Galperin | NCBI, National Library of Medicine, National Institutes of Health, Bethesda, MD, USA |
| Lunch room at Scilifelab, floor 2 | Genomic and biogeochemical clues to the origin of line | ||
|
In the past, origin of life on Earth has been treated mostly as a philosophical problem with little
connection to everyday biological research. Even after the possibility of abiotic origin of amino
acids and nucleic acid bases had been demonstrated in 1953, there has been no agreement on
the energy source(s) for the formation of increasingly complex biopolymers (redox or thermal
gradients, UV, atmospheric electricity, etc.), the driving force(s) leading to the emergence of the
first life forms (natural selection vs spontaneous self-organization), their properties (RNA-based vs
metabolism-based, autotrophic vs heterotrophic, etc.), or place of origin (deep sea vs fresh water).
The availability of genomic data for diverse bacteria, archaea, and eukaryotes, including various
extremophiles, allowed us to take a new look at this problem. By identifying the common genome
core of all (known) living organisms, and the shared properties of their cells, it has become possible
to deduce simple and reasonable biogeochemical constraints on the conditions that led to the origin
of life and to get an insight on where it has happened and how. In turn, these reconstructions lead
to new questions that can now be addressed experimentally, bringing the whole enterprise into the
realm of “normal” science. The most surprising result of these studies is the growing impression that
the origin of life has been a natural consequence of the geochemical conditions that existed on the
primordial Earth, rather than a one-time improbable accident. Mulkidjanian AY and Galperin MY (2009) Biol Direct 4:27. PMID:19703275 Mulkidjanian AY et al. (2012) Proc Natl Acad Sci USA 109:E821. PMID: 22331915 | |||
| Tuesday September 10 | 10:30 | Walter Basile | Stockholm University |
| Lunch room at Scilifelab, floor 2 | Orphan genes | ||
| Title and abstract to come | |||
| Copyright © 2013 Stockholm Bioinformatics Center | any problems? contact the webmaster |