Study On The Protein Structure Prediction Based On Swarm Intelligence Algorithm

The study of protein folding morphology is to determine the regularity,stability,and biologically active structure of proteins under various complex factors.The research basis is how to use the primary structure of the polypeptide chain to obtain the spatial conformation of the protein.In this paper,we focused on the prediction of protein spatial structure and proposed several swarm intelligence algorithms.These methods are based on the folded energy potential well model of proteins,and are derivate algorithms based on the particle swarm optimization and genetic algorithms.Among these methods,the minimum protein potential energy of the spatial structure is the main optimization target.Moreover,we also introduced the amino acids' hydrophobicity and discreteness factors to the main folding energy object,and the protein structure prediction suffered by these objects are also studied.This paper introduces a new algorithm named Potential well Particle Swarm Optimization(PPSO),which is inspired by Brownian movement in nature,especially the microscopic community.With this method,the learning effect of sociology is used to determine the number of particles with the fluctuation feature.To evaluate PPSO,we tested it on six benchmark functions.Our comparative results showed that the PPSO method performs better than the other six state of the art PSO variants.Compared with most PSO variants,the PPSO algorithm is easy to implement and has less control parameters.These attributes of PPSO make it easier to apply in solving real world problems,especially in the prediction of the protein structure.P-PBIL is a method applicable to the integration of genetic searches that can be used for the prediction of the secondary structures of proteins.P-PBIL uses probability vectors to search the protein structure with minimal energy and uses the protein potential well model for particle learning,which requires a much lower machine learning level than the traditional genetic algorithm(GA).P-PBIL lacks any complicated genetic operators and has the characteristics of less computational cost and a large search space.Moreover,P-PBIL algorithm also supports complex allelic structures with good scalability.In this paper,the CB513 and Homo sapiens datasets were used as training sets to obtain the release probability matrices with different amino acid lengths.The test results showed that P-PBIL is superior to current mainstream algorithms in predicting single sequence proteins and it is a better predictive learning method.In this paper,we proposed two hypotheses to explain protein folding.Then,we proposed an improved 3D off-lattice model of the protein folding structure.This model is beneficial to PPSO in greatly reducing the computation complexity so the 3D coordinate of each amino acid can be easily obtained.The particle with spatial search ability is limited to the last amino acid of the peptide chain,and the position of the selected particle is determined by PPSO.The computational studies involved the Fibonacci sequence and the selected real proteins,so the test results verified that the method has strong global searching ability,as it can easily obtain the spatial coordinates of each amino acid in little computational time.The experiment also shows that PPSO has better performance than basic PSO,so it is a feasible solution to this problem.The protein folding process involves many factors,including the force field factor,since the protein folding process is dominated by the force field model.In order to solve the multi-objective optimization problem,a multi-objective protein structure prediction algorithm named Gradient Stochastic Ranking Algorithm(GSRA)was proposed.First,the GSRA assigned the Gaussian niche preservation operation to evaluate the perpendicular distance of every niche member.Second,the GSRA chose two indicators with different biases to balance the convergence and diversity.Comprehensive experimental studies on protein structure prediction cover the energy,hydrophobic and dispersion factors,sensitivity,specificity and the MCC.The test results showed that the GSRA is better than the PPSO algorithm presented in this paper.