BCR CDR3 length distributions differ between blood and spleen and between old and young patients, and TCR distributions can be used to detect myelodysplastic syndrome
BCR CDR3 length distributions differ between blood and spleen and between old and young patients, and TCR distributions can be used to detect myelodysplastic syndrome
复制标题
DOI：
10.1088/1478-3975/10/5/056001
复制
发表时间：
2013-10-01
期刊：
影响因子：
2
通讯作者：
Mehr, Ramit
中科院分区：
文献类型：
--
作者：
Pickman, Yishai;Dunn-Walters, Deborah;Mehr, Ramit
互补决定区3（CDR 3）是B细胞受体（BCR）和T细胞受体（TCR）基因中最高变的区域，并且是抗原识别中最关键的结构，从而决定发育和应答淋巴细胞的命运。人血液中存在数百万种不同的TCR V β链或BCR重链CDR 3序列。即使是现在，当高通量测序变得广泛使用时，CDR 3长度分布（也称为光谱）仍然是评估库多样性的更快和更便宜的方法。然而，分布复杂性和每个样本的大量信息（例如，G. TCR α链的32个分布和TCR β链的24个分布）要求使用机器学习工具进行全面探索。我们已经研究了监督机器学习的能力，它使用计算模型在预定义的生物组中找到隐藏的模式，分析来自各种来源的CDR 3长度分布，并区分实验组。我们发现：（a）与外周血分布相比，脾脏BCR CDR 3长度分布的特征在于低标准差和很少的局部最大值;（B）健康老年人的BCR CDR 3长度分布可以与年轻人的BCR CDR 3长度分布区分开来;（c）基于TCR CDR 3分布特征的机器学习模型可以检测骨髓增生异常综合征，准确率约为93%。总的来说，我们证明了使用监督机器学习方法有助于我们理解淋巴细胞库的多样性。
Complementarity-determining region 3 (CDR3) is the most hyper-variable region in B cell receptor (BCR) and T cell receptor (TCR) genes, and the most critical structure in antigen recognition and thereby in determining the fates of developing and responding lymphocytes. There are millions of different TCR V beta chain or BCR heavy chain CDR3 sequences in human blood. Even now, when high-throughput sequencing becomes widely used, CDR3 length distributions (also called spectratypes) are still a much quicker and cheaper method of assessing repertoire diversity. However, distribution complexity and the large amount of information per sample (e. g. 32 distributions of the TCR alpha chain, and 24 of TCR beta) calls for the use of machine learning tools for full exploration. We have examined the ability of supervised machine learning, which uses computational models to find hidden patterns in predefined biological groups, to analyze CDR3 length distributions from various sources, and distinguish between experimental groups. We found that (a) splenic BCR CDR3 length distributions are characterized by low standard deviations and few local maxima, compared to peripheral blood distributions; (b) healthy elderly people's BCR CDR3 length distributions can be distinguished from those of the young; and (c) a machine learning model based on TCR CDR3 distribution features can detect myelodysplastic syndrome with approximately 93% accuracy. Overall, we demonstrate that using supervised machine learning methods can contribute to our understanding of lymphocyte repertoire diversity.