A pitfall for machine learning methods aiming to predict across cell types.

A pitfall for machine learning methods aiming to predict across cell types.
复制标题

DOI:
10.1186/s13059-020-02177-y
复制
发表时间:
2020-11-19
期刊:
影响因子:
12.3
通讯作者:
Noble WS
Noble WS
中科院分区:
生物学1区
文献类型:
--
作者:
Schreiber J;Singh R;Bilmes J;Noble WS

文献摘要

参考文献

被引文献

相似文献

预测基因组活动的机器学习模型在对不同类型的细胞进行准确预测时最有用。在这里,我们表明,当训练集和测试集包含相同的基因组位置时,通过有效地记住与训练细胞类型中的每个位置相关的平均活动,所得到的模型可能看起来表现良好。我们在预测基因表达和染色质区域边界的背景下证明了这一现象,并提出了诊断和避免陷阱的方法。我们预计,随着更多数据的获得,未来的项目将越来越有可能受到这一问题的影响。网上版载有补充材料,可在网上查阅(doi:10.1186/s13059-020-02177-y)。
Machine learning models that predict genomic activity are most useful when they make accurate predictions across cell types. Here, we show that when the training and test sets contain the same genomic loci, the resulting model may falsely appear to perform well by effectively memorizing the average activity associated with each locus across the training cell types. We demonstrate this phenomenon in the context of predicting gene expression and chromatin domain boundaries, and we suggest methods to diagnose and avoid the pitfall. We anticipate that, as more data becomes available, future projects will increasingly risk suffering from this issue. The online version contains supplementary material available at (doi:10.1186/s13059-020-02177-y).
DOI: 10.1038/nbt.3157
发表时间: 2015-04
影响因子: 46.9
作者:
Ernst J;Kellis M
通讯作者: Kellis M
DOI: 10.1038/s41467-018-03635-9
发表时间: 2018-04-11
影响因子: 16.6
作者:
Durham TJ;Libbrecht MW;Howbert JJ;Bilmes J;Noble WS
通讯作者: Noble WS
DOI: 10.1186/s13059-014-0566-0
发表时间: 2015-01-02
期刊: Genome biology
影响因子: 12.3
作者:
Liu H;Zhang X;Huang J;Chen JQ;Tian D;Hurst LD;Yang S
通讯作者: Yang S
DOI: 10.1093/bioinformatics/btz352
发表时间: 2019-07-15
期刊: BIOINFORMATICS
影响因子: 5.8
作者:
Nair, Surag;Kim, Daniel S.;Kundaje, Anshul
通讯作者: Kundaje, Anshul
DOI: 10.1038/nmeth.1937
发表时间: 2012-03-18
期刊: NATURE METHODS
影响因子: 48
作者:
Hoffman, Michael M.;Buske, Orion J.;Wang, Jie;Weng, Zhiping;Bilmes, Jeff A.;Noble, William Stafford
通讯作者: Noble, William Stafford