IIBR Informatics: Advancing Bioinformatics Methods using Ensembles of Profile Hidden Markov Models

生物研究管道中的许多步骤都涉及到机器学习模型的使用，这些模型已经成为许多基本问题的标准工具。与标准用法相比，对基本机器学习模型（机器学习模型的“集成”）的阐述可以提高各种生物学问题的准确性。然而，这些集成的设计是相当特别的，并且它们的使用可能是计算密集型的，这降低了它们在实践中的吸引力。该项目将通过开发用于构建机器学习模型集合的统计严格技术来推进这项技术，以提高准确性。该项目还将开发使用这些集成的方法来解决新的生物学问题，包括蛋白质结构和功能预测。更广泛的影响包括软件学校，与弱势群体的接触，以及开源软件。隐马尔可夫模型（Profile Hidden Markov Models，即Profile hmm）是生物信息学中广泛应用的概率图形模型。过去十年的研究表明，在生物信息学的许多应用中，包括系统发育定位、多序列比对和宏基因组读段的分类鉴定，基因组HMM集成（e-HMM）比单个HMM提供更高的准确性。该项目将通过开发严格的统计技术来构建e- hmm，以提高准确性和增进对e- hmm的理解，从而推进e- hmm的使用，并将开发使用e- hmm进行蛋白质结构和功能预测的方法。更广泛的影响包括软件学校，与代表性不足的群体的接触，以及开源软件。项目软件和论文可在http://tandy.cs.illinois.edu/eHMMproject.html.This上获得，该奖项反映了NSF的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Many steps in biological research pipelines involve the use of machine learning models, and these have become standard tools for many basic problems. Elaborations on basic machine learning models ("ensembles" of machine learning models) can provide improvements in accuracy compared to standard usage, for various biological questions. However, the design of these ensembles has been fairly ad hoc, and their use can be computationally intensive, which reduces their appeal in practice. This project will advance this technology by developing statistically rigorous techniques for building ensembles of machine learning models, with the goal of improving accuracy. The project will also develop methods that use these ensembles for new biological problems, including protein structure and function prediction. Broader impacts include software school, engagement with under-represented groups, and open-source software. Profile Hidden Markov Models (i.e., profile HMMs) are probabilistic graphical models that are in wide use in bioinformatics. Research over the last decade has shown that ensembles of profile HMMs (e-HMMs) can provide greater accuracy than a single profile HMM for many applications in bioinformatics, including phylogenetic placement, multiple sequence alignment, and taxonomic identification of metagenomic reads. This project will advance the use of e-HMMs by developing statistically rigorous techniques for building e-HMMs with the goal of improving accuracy and improving understanding of e-HMMs, and will also develop methods that use e-HMMs for protein structure and function prediction. Broader impacts include software schools, engagement with under-represented groups, and open-source software. Project software and papers are available at http://tandy.cs.illinois.edu/eHMMproject.html.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.