Mingyan Jiang

Reverse engineering of gene regulatory networks using flexible neural tree models

The advances on DNA microarray technologies have enabled researchers to gain hundreds to thousands of gene expression levels. Much effect has been devoted over the past decade to analyze the gene expression data. In this study, flexible neural tree (FNT) model is used for gene regulatory network reconstruction and time-series prediction from gene expression profiling. We use voting strategy and Akaike information criterion (AIC) as two methods to identifying minimal regulatory elements of a target gene. A simulated dataset and three real biological datasets are used to test the validity of the FNT model. Results reveal that the FNT model can improve the prediction accuracy of microarray time-series data effectively and reconstruct gene regulatory network accurately.

Ensemble of Flexible Neural Tree and Ordinary Differential Equations for Small-Time Scale Network Traffic Prediction

Accurate models play important roles in capturing the salient characteristics of the network traffic, analyzing and simulating for the network dynamic, and improving the predictive ability for system dynamics. In this study, the ensemble of the flexible neural tree (FNT) and system models expressed by the ordinary differential equations (ODEs) is proposed to further improve the accuracy of time series forecasting. Firstly, the additive tree model is introduced to represent more precisely ODEs for the network dynamics. Secondly, the structures and parameters of FNT and the additive tree model are optimized based on the Genetic Programming (GP) and the Particle Swarm Optimization algorithm (PSO). Finally, the expected level of performance is verified by using the proposed method, which provides a reliable forecast model for small-time scale network traffic. Experimental results reveal that the proposed method is able to estimate the small-time scale network traffic measurement data with decent accuracy.

A fast and efficient method for inferring structure and parameters of S-system models

Most previous methods of inferring the S-system models have a significant limitation. That is, the structure of S-system models is fixed, and the only goal is to optimize its parameters and coefficients. Because the number of S-system parameters is proportional to the square of the number of variable, a large number of S-system parameters need to be simultaneously estimated, when the number of variable is very large. This limit may lead to the overwhelming computational complexities. To overcome this limitation we propose the restricted additive tree model for inferring the S-system models. In this approach, the evolution algorithm based on tree-structure and the particle swarm optimization (PSO) are employed to evolve the structure and the parameters of the S-system models, respectively. And the partitioning strategy is used to reduce the search space. We make three experiments and Simulation results using both synthetic data and real microarray measurements show that the structures and parameters of the S-system models can be identified correctly, which demonstrate the effectiveness of the proposed methods. The experiment results show that the structures and parameters of the S-system models can be identified correctly. And compared with other methods, the spent time is sharply reduced.

Inferring gene regulatory networks using a time-delayed mass action model.

This paper demonstrates a new time-delayed mass action model which applies a set of delay differential equations (DDEs) to represent the dynamics of gene regulatory networks (GRNs). The mass action model is a classical model which is often used to describe the kinetics of biochemical processes, so it is fit for GRN modeling. The ability to incorporate time-delayed parameters in this model enables different time delays of interaction between genes. Moreover, an efficient learning method which employs population-based incremental learning (PBIL) algorithm and trigonometric differential evolution (TDE) algorithm TDE is proposed to automatically evolve the structure of the network and infer the optimal parameters from observed time-series gene expression data. Experiments on three well-known motifs of GRN and a real budding yeast cell cycle network show that the proposal can not only successfully infer the network structure and parameters but also has a strong anti-noise ability. Compared with other works, this method also has a great improvement in performances.

Adaptive Sub-Channel Allocation Based on Hopfield Neural Network for Multiuser OFDM

A kind of adaptive sub-channel allocation method utilizing Hopfield neural network (HNN) is studied in this paper. In order to find the power optimal sub-channel allocation under the constraints that only one sub-channel can be allocated to one user and all users are allocated the same number of sub-channels, a kind of new energy constrained function is constructed for the HNN. It is shown through numerical simulation that the proposed method can find the optimal allocation with less complexity compared with the exhaustive method.

A novel ensemble of probabilistic neural network for predicting protein-protein interactions

The knowledge of protein-protein interactions (PPIs) in cells is indispensable for deep understanding the biological process. Although many computational methods have been developed for identification of PPIs, there are still many difficulties due to high computation complexity and noisy data. In this paper, we proposed an ensemble of probabilistic neural network (PNN) to predict PPIs from primary sequence which achieved promising results. The key advantage of the algorithm is that it combines variety of physicochemical property features to construct diverse individual classifiers for ensemble prediction. What makes the method much more attractive is that it not only generated much more diverse and robust individual classifiers, but also contains different interaction physicochemical information which dictated the structure and the function of proteins. Moreover, the PNN is robust to noise and trained easily, it is suitable for dealing with the large scale noisy PPIs data. Experiment results on H. pylori and Human datasets show that our proposed method performs at least 8% higher accuracy than the best of other related works.