SGER: Belief Networks for Human Pathways

对复杂对象的分析，如人类精神障碍，目前被分成多个科学社区，这些社区通常很少或根本没有互动。如果将不同的知识拼图组合在一个脑袋里，很有可能，甚至很有可能，现在就可以找到找到治疗无数人类疾病的药物的有力线索。更可取的是，用定量的概率模型取代无序的事实收集，以便对模型预测进行正式评估，分析数据点之间的差异，并进行假设检验。长期研究计划将解决许多问题，例如汇编数据，将互动转化为信念，以及放大与每个节点相关的数据。对这一设想的可行性的测试始于一个众所周知的系统，即面包师酵母中的细胞周期网络，以信念网络形式主义复制已知的酵母表型效应。这项分析的目标是(1)定义应用于路径数据的信念网络方法的适用范围，(2)证明该方法的可行性，以及(3)将所得到的模型用作更大规模研究的原则证明。下一步将包括用信念网络对几年来汇编的关于人类自闭症的知识进行建模，重点是将分子相互作用数据自动转换为信念网络，计算个人相互作用的概率，纳入专家输入，以及通过网络进行推理的实验。私人助理和一名研究生将参与这一项目。

ABSTRACTNSF-0438921RZHETSKY, ANDREYAnalysis of complex subjects, such as human psychiatric disorders, is currently fragmented into multiple scientific communities, which often have little or none interactions. It is quite possible, even likely, that powerful cues to finding remedies to numerous human maladies can be found right now provided that disparate pieces of the knowledge puzzle are combined in one head. It would be even more desirable to have the unordered collection of facts substituted with a quantitative probabilistic model allowing for formal evaluations of model predictions, analysis of discrepancies between data points, and hypothesis testing. The long-term research plan would address many issues, such as compiling the data, converting interactions into beliefs and amplifying the data associated with each node. The test of the feasibility of this vision begins with a well-known system, cell-cycle network in the baker.s yeast to reproduce with belief network formalism the known phenotypic effects for yeast. The goals in this analysis would be (1) define applicability boundaries of the belief network methodology as applied to pathway data, (2) demonstrate feasibility of the approach, and (3) use the resulting model as a proof-of-principle for a larger study. The next step would include modeling with belief networks the knowledge that has been compiled over a few years on autism in humans, with a focus on automated conversion of molecular interaction data into belief networks, computation of probabilities for individual interactions, incorporation of expert inputs and experiments with reasoning over the network. The PI and a graduate student will be engaged in this project.