Nonparametirc Bayes-based infinite mixture model algorithms for Bioinformatics

我们提出了一种非参数贝叶斯模型用于两个生物信息学应用：1)蛋白质功能自动预测和2)基因表达网络推断。对于蛋白质功能的自动预测，我们提出了一种新的预测蛋白质功能的方法，称为PreGO。PreGO是一种基于无限混合隐马尔可夫模型的算法。给定一个未注释的蛋白质序列，PreGO预测基因本体术语存在的概率。对于基因表达数据的时变网络推断，我们采用非参数贝叶斯回归方法来预测基因之间的相互作用。该方法有望在时变网络中实现更灵活的回归能力。为了获得对噪声数据更强的鲁棒性，我们采用了t过程。基本算法采用可逆跳跃马尔可夫链蒙特卡罗对整个网络结构进行推理。该方法可以同时处理(1)变化点检测和(2)网络结构推断。

We proposed a non-parametric Bayesan models to two bioinformatics applications: 1) automatic protein function prediction and 2) gene expression network inference. For automatic protein function prediction,we proposed a novel method to predict protein functions, called PreGO. PreGO is an algorithm based on an infinite mixture of hidden Markov models. Given an unannotated protein sequence, PreGO predicts the probability of existence of Gene Ontology terms. For time-varying network inference for gene expression data, we adopted a nonparametric Bayesian regression method to predict interactions between the genes. This method is expected to achieve more flexible regression capability in time-varying network. To obtain stronger robustness to noisy data, we employed the T-Process. The basic algorithm employed reversible jump Markov Chain Monte Carlo for inference of whole network structures. The method can handle (i) change point detection and (ii) network structure inference simultaneously.