Department of Neuroimaging, Institute of Psychiatry, Psychology and Neuroscience, King's College London, London, UK.
Invicro, W12 0NN, London, UK; Division of Brain Sciences, Department of Medicine, Imperial College London, SW72AZ, London, UK.
J Neurosci Methods. 2021 May 1;355:109128. doi: 10.1016/j.jneumeth.2021.109128. Epub 2021 Mar 17.
The Allen Human Brain Atlas (AHBA) is the first example of human brain transcriptomic mappings and detailed anatomical annotations which, for the first time, has allowed the integration of human brain transcriptomics with neuroimaging. This has been made possible because the AHBA offered an original dataset that contains mRNA expression measures for >20,000 genes covering the whole brain and, critically, in a standard stereotaxic space. In recent years many different methods have been used to integrate this data set with brain imaging data, although this endeavour has lacked harmony in terms of the workflow of data processing and subsequent analyses. In this work we discuss five main issues that experience has highlighted as in need of thorough consideration when integrating the AHBA with neuroimaging. These concerns are corroborated by comparing the performance of three different publicly available methods in correlating the same measures of serotonin receptors density with the correspondent AHBA mRNA maps. In this representative case, we were able to show how these methods can lead to discrepant results, suggesting that processing options are not neutral. We believe that the field should take into serious consideration these issues as they could undermine reproducibility and, in the end, the intrinsic value of the AHBA. We also advise on possible strategies to overcome these discrepancies. Finally, we encourage authors towards practices that will improve reproducibility such as transparency in reporting, algorithm and data sharing, collaboration.
艾伦人类大脑图谱 (Allen Human Brain Atlas, AHBA) 是人类大脑转录组图谱和详细解剖注释的首个范例，它首次实现了人类大脑转录组学与神经影像学的整合。这之所以成为可能，是因为 AHBA 提供了一个原始数据集，其中包含了覆盖整个大脑的 >20000 个基因的 mRNA 表达测量值，而且关键的是，这些数据是在标准的立体空间中获得的。近年来，已经使用了许多不同的方法将这个数据集与脑成像数据进行整合，尽管在数据处理和后续分析的工作流程方面缺乏协调性。在这项工作中，我们讨论了五个主要问题，这些问题是在将 AHBA 与神经影像学进行整合时需要彻底考虑的，这些问题通过比较三种不同的、可公开获取的方法在将相同的 5-羟色胺受体密度测量值与相应的 AHBA mRNA 图谱相关联时的性能得到了证实。在这个有代表性的案例中，我们能够展示这些方法如何导致不一致的结果，这表明处理选项并不是中立的。我们认为，该领域应该认真考虑这些问题，因为它们可能会破坏可重复性，最终破坏 AHBA 的内在价值。我们还就克服这些差异的可能策略提供了建议。最后，我们鼓励作者采取提高可重复性的实践，如报告、算法和数据共享的透明度、合作。
