Sulfate reduction is an essential metabolism that maintains biogeochemical cycles in marine and terrestrial ecosystems.Sulfate reducers are exclusively prokaryotic,phylogenetically diverse,and may have evolved early in Earth’s history.However,their origin is elusive and unequivocal fossils are lacking.Here we report a new microfossil,Qingjiangonema cambria,from518-million-year-old black shales that yield the Qingjiang biota.Qingjiangonema is a long filamentous form comprising hundreds of cells filled by equimorphic and equidimensional pyrite microcrystals with a light sulfur isotope composition.Multiple lines of evidence indicate Qingjiangonema was a sulfate-reducing bacterium that exhibits similar patterns of cell organization to filamentous forms within the phylum Desulfobacterota,including the sulfate-reducing Desulfonema and sulfide-oxidizing cable bacteria.Phylogenomic analyses confirm separate,independent origins of multicellularity in Desulfonema and in cable bacteria.Molecular clock analyses infer that the Desulfobacterota,which encompass a majority of sulfate-reducing taxa,diverged~2.41 billion years ago during the Paleoproterozoic Great Oxygenation Event,while cable bacteria diverged~0.56 billion years ago during or immediately after the Neoproterozoic Oxygenation Event.Taken together,we interpret Qingjiangonema as a multicellular sulfate-reducing microfossil and propose that cable bacteria evolved from a multicellular filamentous sulfate-reducing ancestor.We infer that the diversification of the Desulfobacterota and the origin of cable bacteria may have been responses to oxygenation events in Earth’s history.
软件测试在软件安全保障和质量保证流程中扮演着关键角色,为了降低软件的维护成本,提高软件的安全性,需要尽早地发现和修复漏洞和问题。因此在软件版本迭代的过程中,测试代码需要在生产代码修改后及时更新。然而,测试代码往往难以和生产代码同步更新,导致测试效果不佳。协同演化方法被用于解决这一问题,但维持这一模式的成本较高。本文分析了现有研究,利用关联规则挖掘技术,研究了生产代码和测试代码之间的协同演化关系,提出了一种基于神经机器翻译(Neural Machine Translation,NMT)的测试用例协同演化方法(NMT-based Test Case Co-evolution,NTCC)。通过分析生产代码和测试代码的历史提交来提取测试用例的特征,在历史提交的生产代码和测试上进行训练,然后在一个较小的生产代码数据集上进行调整,最后通过集束搜索的方法产生测试代码。实验结果表明,NTCC方法可以有效地识别生产-测试代码协同演化,准确率达到了78.33%,在正类和负类上的F1-分数分别为80.10%和76.22%,优于基线方法。