Study on Modelling Protein LOOP Structure by Bayesian Network
|Course||Applied Computer Technology|
|Keywords||computational modeling for protein loop Bayesian network structurelearning bivariate von Mises distribution|
Modeling protein loops is a hard unsolved problem of computational biology, butis important for understanding characteristics and functions of protein. To obtain the3D structure of the protein loops with experimental methods is challenging because ofloops’high flexibility. Hence, to model the loops’structure as accurate as possible bycomputing is critical for determining the complete protein structure.By employing a general Bayesian network, this paper constructs a fully proba-bilistic continuous model of protein loops, where the continuous torsion angle pairs ofthe loops are represented by bivariate von Mises distribution. In this model, loops’amino acid information as the torsion angle pair’s causal variables explicitly controlthe loops’structure. In order to evaluate the feasibility and eflectiveness of continuousmodeling for loop structure, this paper choose the bivariate von Mises distributionto construct a LoopMM model, whose structure is fixed. After learning the trainingset, extracted the loops part from SABmark, LoopMM can sample continuous torsionangle pairs. The results of experimenting on the eight free modeling targets of CASP8show that the torsion angle pairs sampled by LoopMM are more close to the nativetorsion angle pair than other methods’.A new Bayesian network is further learned from the same training set but by re-leasing the fixed structure of LoopMM. This network diflers from LoopMM by allowingsome long-distance residues’s amino acid combined with the secondary structure to de-termine another residues’s torsion angle pair. The new model is called LoopBN whichcan also sample the torsion angle pairs with more accuracy and helps to improve thefull protein backbone accuracy for Abinitio prediction method.Modelling protein loops structure by Bayesian Network is valuable in two aspects.On the one hand it provides a new computational model to improve the accuracy ofstructure prediction. On the other hand it can also describe some potential causal-ity among the residues of loops. The relationship between diflerent residues’aminoacids or secondary structures and the values of torsion angle pair can be showed in an intuitionistic way with the directed edges in the Bayesian network. Such a computa-tional approach helps to discover the scientific insights in a more understandable andinterpretable way for biological problems.