基于基因扰动及变分逼近技术的基因调控网络推断

Variational approximation inference for gene regulatory networks from gene perturbations

为了有效提高基因调控网络推断的精度,基于基因表达数据和基因扰动数据,将基因调控网络建模为结构方程模型,并进一步转化为验证性因子分析(CFA)模型,然后使用贝叶斯方法求解CFA模型参数.在贝叶斯分析中,为减少计算量,不采用常用的马尔科夫-蒙特卡洛抽样方法,而是采用变分逼近技术对参数的联合后验分布进行因式化,并获得参数的后验包含概率分布及参数的后验分布.同时使用重要性抽样技术对CFA模型的推断参数进行加权平均.仿真结果表明,CFA模型和变分逼近技术是有效和可靠的.根据实验数据,使用所提算法推导了具有35个基因的酵母基因调控网络.

To improve the inference accuracy of gene regulatory networks （GRN）,using both gene perturbations and gene expression data,GRN is modeled as a structural equation model （SEM）,and further transformed into the confirmatory factor analysis （CFA）model.The Bayesian approach is used to infer the parameters of the regulatory networks.Instead of the Markov chain Monte Carlo （MCMC）method,the variational approximation method （VAM）is applied for its lower computa-tion cost,which factorizes the joint posterior distribution of parameters,and obtains the posterior in-clusion probability distribution and the posterior distribution of parameters.An importance sampling technique is then applied to obtain the weighted average of the CFA inferred parameters.Simulations are carried out to verify the effectiveness and reliability of the CFA model and the variational approx-imation.Based on the experimental data,the regulatory interactions among a 35 yeast genes network are identified with the proposed VAM algorithm.