DBN FOLD RECOGNITION · Structure based fold recognition using Dynamic Bayesian Networks
FP6 — Marie Curie Actions (Human Resources and Mobility)
- Duration
- 2004-06-01 → 2006-05-31
- EU contribution
- €182,737
- Participants
- 1
- Scheme
- EIF
Lines connect the coordinator with its partners. CORDIS does not always give exact coordinates for projects before 2014. These points are placed at city or country level.
Results in brief
Final Activity Report Summary - DBN FOLD RECOGNITION (Structure based fold recognition using Dynamic Bayesian Networks)
The prediction of protein structure from sequence is without doubt one of the most important open problems in computational biology, and biology in general. Currently, the most successful prediction methods use a divide-and-conquer approach to attack the problem. Typically, protein-like conformations are generated and evaluated using very approximate energy functions. After the generation of a large number of candidates, the putative native-like conformations is somehow selected, most often by clustering and the use of an expensive all-atom energy function. One of the main bottlenecks in the process is the generation of protein-like conformations, or, in other words, the exploration of the conformational landscape. Currently, this is typically done by tying together discrete fragments from existing structures. However, this approach has many disadvantages: it discretizes the continuous conformational space, is prone to data sparseness and cannot be used in a probabilistically sound way. We developed a probabilistic model of the local structure of proteins, based on dynamic Bayesian networks (DBNs) and directional statistics that solves these problems. The model allows sampling of protein-like conformations in continuous space, and is computationally efficient, mathematically rigorous, probabilistic and conceptually elegant. We expect this model will lead to important breakthroughs in de novo structure prediction, fold recognition, protein design and experimental determination of protein structure. The method was published in the journal PLoS Computational Biology, and featured on the cover of the September 2006 issue. In addition, we solved an important subproblem that arises in the use of the model of local structure for the above mentioned applications (loop closure in C-alpha space).
Data: CORDIS, © European Union
Project objective
Structure Based Protein Fold Recognition Using Dynamic Bayesian Networks Today it remains impossible to predict the three-dimensional structure of a protein based on its amino acid sequence alone. The ability to predict protein structure from sequence would have great impact on crucial areas ranging from understanding protein function over protein design to drug development. To simplify the problem, fold recognition methods try to evaluate the possibility that a query sequence adopts any of the already known folds (the template folds). This is already extremely useful, since knowing that a sequence likely corresponds to a certain fold can e.g. lead to the elucidation of the protein's function. This can e.g. be used to perform gene annotation on a genomic scale. We propose to construct a novel structure based protein fold recognition method that uses a probabilistic description of protein structure using Dynamic Bayesian Networks (Dens), a machine learning method for which efficient methods exist to perform inference and parameter learning from data. This will allow us to construct probabilistic models that are in many ways more advanced than currently used techniques. In particular, we will develop Dens that represents local amino acid preference, local backbone structure, residue exposure and no local residue-residue contacts of a 'general' protein structure. These probabilistic models can then be used to evaluate the fit between a given sequence family superimposed on a specific structure, le. As a scoring function in a fold recognition method.
Original text from CORDIS.
Participants
- KOEBENHAVNS UNIVERSITET · COPENHAGEN KCoordinatorCity levelDenmark
Links
Data: CORDIS, © European Union