Praveen Koduru

A Multiobjective Evolutionary-Simplex Hybrid Approach for the Optimization of Differential Equation Models of Gene Networks

This paper describes genetic and hybrid approaches for multiobjective optimization using a numerical measure called fuzzy dominance. Fuzzy dominance is used when implementing tournament selection within the genetic algorithm (GA). In the hybrid version, it is also used to carry out a Nelder-Mead simplex-based local search. The proposed GA is shown to perform better than NSGA-II and SPEA-2 on standard benchmarks, as well as for the optimization of a genetic model for flowering time control in rice. Adding the local search achieves faster convergence, an important feature in computationally intensive optimization of gene networks. The hybrid version also compares well with ParEGO on a few other benchmarks. The proposed hybrid algorithm is then applied to estimate the parameters of an elaborate gene network model of flowering time control in Arabidopsis. Overall solution quality is quite good by biological standards. Tradeoffs are discussed between accuracy in gene activity levels versus in the plant traits that they influence. These tradeoffs suggest that data mining the Pareto front may be useful in bioinformatics.

Fuzzy Dominance Based Multi-objective GA-Simplex Hybrid Algorithms Applied to Gene Network Models

Hybrid algorithms that combine genetic algorithms with the Nelder-Mead simplex algorithm have been effective in solving certain optimization problems. In this article, we apply a similar technique to estimate the parameters of a gene regulatory network for flowering time control in rice. The algorithm minimizes the difference between the model behavior and real world data. Because of the nature of the data, a multi-objective approach is necessary. The concept of fuzzy dominance is introduced, and a multi-objective simplex algorithm based on this concept is proposed as a part of the hybrid approach. Results suggest that the proposed method performs well in estimating the model parameters.

Adding Local Search to Particle Swarm Optimization

Particle swarm optimization is a stochastic algorithm for optimizing continuous functions. It uses a population of particles that follow trajectories through the search space towards good optima. This paper proposes adding a local search component to PSO to improve its convergence speed. Two possible methods are discussed. The first adds a term containing estimated gradient information to the velocity of each particle. The second explicitly incorporates the Nelder-Mead algorithm, a known local search technique, within PSO. The suggested methods have been applied to the problem of estimating parameters of a gene network model. Results indicate the effectiveness of the proposed strategies.

Multi-objective hybrid PSO using µ-fuzzy dominance

This paper describes a PSO-Nelder Mead Simplex hybrid multi-objective optimization algorithm based on a numerical metric called µ -fuzzy dominance. Within each iteration of this approach, in addition to the position and velocity update of each particle using PSO, the k-means algorithm is applied to divide the population into smaller sized clusters. The Nelder-Mead simplex algorithm is used separately within each cluster for added local search. The proposed algorithm is shown to perform better than MOPSO on several test problems as well as for the optimization of a genetic model for flowering time control in Arabidopsis. Adding the local search achieves faster convergence, an important feature in computationally intensive optimization of gene networks.

The gene regulatory network: an application to optimal coverage in sensor networks

This paper proposes a new approach for biologically inspired computing on the basis of Gene Regulatory Networks. These networks are models of genes and dynamic interactions that take place between them. The differential equation representations of such networks resemble neural networks as well as idiotypic networks in immune system. Although several potential applications have been outlined, an example, the problem of placing sensors optimally in a distributed environment is considered in detail. A comparison with NSGA-II suggest that the new method is able to accomplish near-optimal coverage of sensors in a network.

A multi-objective GA-simplex hybrid approach for gene regulatory network models

Genetic algorithms are exploratory search techniques that rely on a large population of individuals. In order to improve the search process, several hybrid approaches have been proposed that make use of a local exploitative search technique, such as the Nelder-Mead simplex algorithm. The simplex algorithm has been modified for multi-objective optimization, by introducing the concept of fuzzy dominance. A hybrid algorithm, the fuzzy dominance based simplex - genetic algorithm (FSGA) has been proposed. This algorithm was shown to be a very effective search strategy when applied to a multi-objective problem in modelling the gene regulatory network of flowering time control of Oryza sativa.

A co-evolutionary hybrid algorithm for multi-objective optimization of gene regulatory network models

In this paper, the parameters of a genetic network for rice flowering time control have been estimated using a multi-objective genetic algorithm approach. We have modified the recently introduced concept of fuzzy dominance to hybridize the well-known Nelder Mead Simplex algorithm for better exploitation with a multi-objective genetic algorithm. A co-evolutionary approach is proposed to adapt the fuzzy dominance parameters. Additional changes to the previous approach have also been incorporated here for faster convergence, including elitism. Our results suggest that this hybrid algorithm performs significantly better than NSGA-II, a standard algorithm for multi-objective optimization.

Simultaneous structure discovery and parameter estimation in gene networks using a multi-objective GP-PSO hybrid approach

This paper presents a hybrid algorithm based on Genetic Programming (GP) and Particle Swarm Optimisation (PSO) for the automated recovery of gene network structure. It uses gene expression time series data as well as phenotypic data pertaining to plant flowering time as its input data. The algorithm then attempts to discover simple structures to approximate the plant gene regulatory networks that produce model gene expressions and flowering times that closely resemble the input data. To show the efficacy of the proposed approach, simulation results applied to flowering time control in Arabidopsis thaliana are demonstrated and discussed.

A particle swarm optimization approach for estimating parameter confidence regions

Point estimates of the parameters in real world models convey valuable information about the actual system. However, parameter comparisons and/or statistical inference requires determination of parameter space confidence regions in addition to point estimates. In most practical applications, the relation of the parameters to model fitness is highly nonlinear and noisy data leads to further deviations. Thus the confidence regions obtained by using locally linearized models are often misleading. Uniform covering by probabilistic rejection (UCPR) is a robust technique that has been developed to solve this problem, and has been proven to be more efficient than other approximate random search techniques. In this paper, we propose a contour particle swarm optimization (C-PSO) technique and compare its performance against UCPR in predicting the confidence regions. Results indicate that for problems with low number of parameters, both the algorithms are quite comparable. However, real world models such as genetic networks have a large number of parameters and the UCPR fails in finding good convergence due to its limited search capabilities. In such problems, the C-PSO technique was able to find the confidence regions with better resolution and efficiency.

Discovering structures in gene regulatory networks using genetic programming and particle swarms

In this paper, we describe a Genetic Programming and Particle Swarm Hybrid algorithm for Gene Network discovery.