Protein prediction represents one of the forefront challenges within bioinformatics due to its computational dificulties. We utilise decoy-based modelling to find a good protein model in the low energy region of the protein energy... more
Protein prediction represents one of the forefront challenges within bioinformatics due to its computational dificulties. We utilise decoy-based modelling to find a good protein model in the low energy region of the protein energy landscape, by stablishing a three-dimensional space via SVD where sampling and optimization is performed via PSO. The goal is to find a representative protein backbone structure and its uncertainty. Current Challenges Theoretical Background Computational Experiments – 2l3f Protein Future Work We could remark the following challenges faces within the field: • Experimental methods used to determine protein structures are very costly. • Good physical models of protein energy and interactions are required to mimic the reality. • Current algorithms are not capable of sampling every single plausible conformation since simplifications exist. • Current computational approaches are inefficient and computationally expensive. • The algorithm is described as follows: • The utilization of SVD allows optimization over a reduced space until the backbone structure found meets the criterion í µí°¦ − í µí°¦ í µí² í µí°»í µí°¸í µí°¦ í µí² í µí°¦ − í µí°¦ í µí² ≤ í µí°¸− í µí°¸í µí°¦ í µí² in a linear hyper-quadratic area. • We presented a fast and simple model redcution technique to predict protein tertiary structures that is capable of reducing the ill-posed character of this highly-dimensional problema. • Other posible reductions techniques such as Disrete Cosine Transformation is about to be explored. • Local optimization techniques and the integration of PSO in Rosetta server are about to be performed.
