Next generation phylogenetic modelling using machine learning

生物遗传学的目标是从现在的多样性中推断过去。 虽然系统发生学最为人所知的是它在重建生物历史（从生物多样性）方面的应用，但系统发生学也对重建语言历史（从世界语言多样性）的问题产生了影响。 我提出的计划包括生物和语言重建。 我假设这两个问题有足够的相似之处，以证明联合研究，但也有足够的差异，以促进创新。 在生物学和语言学中，系统发育研究的现状令人振奋。 研究人员越来越多地拥有新型数据集，即新的数据源，例如语言记录（类型学或语音学）、地理位置、SNP研究结果以及这些的组合。 研究人员也越来越多地获得新规模的数据集，即包含大量物种/种群/语言（统称为分类群）的数据集，或每个分类群的大量数据（全基因组和全词汇研究），或两者兼而有之。 有效地使用新类型和规模的系统发育数据集在很大程度上仍然是一个悬而未决的问题。 这是一个重要的问题，因为它有可能促进我们对基本科学领域的理解，并对生物技术产生影响。 该计划的目的是释放新类型和规模的系统发育数据集的潜力。 该方法将应用并建立在机器学习的最新发展基础上。 由于系统发育推断具有挑战性的特征，这些特征仅在机器学习的前沿发现，这不仅可能导致遗传学的创新，而且可能导致机器学习的创新。

The goal of phylogenetics is to draw inferences about the past from the diversity of the present. Although phylogenetics is best known from its application to the reconstruction of biological histories (from biodiversity), phylogenetic also has ramifications for the problem of reconstructing linguistic histories (from the world's linguistic diversity). The program I propose involves both biological and linguistic reconstructions. I hypothesize that the two problems have sufficient similarities to justify a joint study, but also have sufficient differences to foster innovation. In both biology and linguistics, the current state of phylogenetic research is stimulating. Increasingly, researchers have at their disposal datasets of new types, that is, new sources of data such as linguistic records (typological or phonological), geographic locations, results from SNPs studies, and combinations of these. Increasingly, researcher also have access to datasets of new scales, that is, datasets containing large numbers of species/populations/languages (collectively called taxa), or large amounts of data for each taxon (genome-wide and vocabulary-wide studies), or both. Efficiently using phylogenetic datasets of new types and scales is still largely an open problem. It is an important problem, since it has the potential to advance our understanding in fundamental areas of science and also to impact biotechnology. The aim of this program is to unlock the potential of phylogenetic datasets of new types and scales. The approach will be to apply and build on recent developments in machine learning. Since phylogenetic inference has challenging characteristics that are only found at the cutting edge of machine learning, this is likely to result not only in innovations in phylogenetics, but also in machine learning.