Access through your institution

通过您的机构访问

Buy or subscribe

购买或订阅

Many computational approaches have been developed to predict gene expression level, a single numerical value summarizing the expression profile of a gene. Despite its practical convenience, this simplified view fails to account for the full range of complexities involved in gene expression, such as gene structure, splicing and polyadenylation.

已经开发了许多计算方法来预测基因表达水平，即总结基因表达谱的单个数值。尽管它具有实际的便利性，但这种简化的观点未能解释基因表达所涉及的全部复杂性，例如基因结构，剪接和聚腺苷酸化。

To tackle this limitation, Johannes Linder and David Kelley, both from Calico Life Sciences, and their colleagues built the model Borzoi to predict RNA-seq coverage from DNA sequences..

为了解决这个限制，来自Calico Life Sciences的Johannes Linder和David Kelley及其同事建立了Borzoi模型，以预测DNA序列的RNA-seq覆盖率。。

Borzoi leverages the core architecture of the Enformer model previously developed by the team for predicting gene expression levels. To model RNA-seq coverage spanning the whole gene potentially regulated by various proximal and distal sequence elements, the team use a number of modelling strategies to both increase the sequence length to >500 kb (2.5× larger than Enformer) to cover more gene spans and decrease the coverage track bin size to 32 bp (4× smaller than Enformer) to provide more precision around exon boundaries, notes Kelley.

Borzoi利用了该团队先前开发的用于预测基因表达水平的Enformer模型的核心架构。凯利注意到，为了模拟可能受各种近端和远端序列元件调控的整个基因的RNA-seq覆盖率，该团队使用了许多建模策略，将序列长度增加到>500 kb（比Enformer大2.5倍），以覆盖更多的基因跨度，并将覆盖轨道箱大小减小到32 bp（比Enformer小4倍），以在外显子边界周围提供更高的精度。

Although biologically appealing, this model scale poses computational challenges — for example, for the self-attention neural network blocks, whose memory scales quadratically with sequence length, comments Kelley. “To make it work, we borrowed a technique from image analysis called U-net where we perform self-attention at 128-bp resolution and then zoom back in to 32 bp using U-net skip connections from the initial convolution tower.”.

Kelley评论说，尽管在生物学上很有吸引力，但这种模型规模带来了计算上的挑战，例如，对于自我注意神经网络块，其记忆与序列长度呈二次关系。“为了使它发挥作用，我们从图像分析中借用了一种称为U-net的技术，我们以128 bp的分辨率进行自我注意，然后使用来自初始卷积塔的U-net跳过连接将其放大到32 bp。”。

This is a preview of subscription content,

这是订阅内容的预览，

access via your institution

通过您的机构访问

Access options

访问选项

Access through your institution

通过您的机构访问

Access through your institution

通过您的机构访问

Change institution

变革机构

Buy or subscribe

购买或订阅

Access Nature and 54 other Nature Portfolio journals

Access Nature和54种其他Nature投资组合期刊

Get Nature+, our best-value online-access subscription

获取Nature+，我们最具价值的在线访问订阅

24,99 €

24,99 €

/ 30 days

/30天

cancel any time

随时取消

Learn more

了解更多信息

Subscription info for Chinese customers

中国客户的订阅信息

We have a dedicated website for our Chinese customers. Please go to

我们有一个专门为中国客户服务的网站。请转到

naturechina.com

naturechina.com

to subscribe to this journal.

订阅此日记。

Go to naturechina.com

访问naturechina.com

Buy this article

购买这篇文章

Purchase on SpringerLink

在SpringerLink上购买

Instant access to full article PDF

即时访问全文PDF

Buy now

立即购买

Prices may be subject to local taxes which are calculated during checkout

价格可能需要缴纳结帐时计算的地方税

Additional access options:

其他访问选项：

Log in

登录

Learn about institutional subscriptions

了解机构订阅

Read our FAQs

阅读我们的常见问题

Contact customer support

联系客户支持

Author information

作者信息

Authors and Affiliations

作者和隶属关系

Nature Methods

自然方法

https://www.nature.com/nmeth/

https://www.nature.com/nmeth/

Lin Tang

林堂（音）

Authors

作者

Lin Tang

林堂（音）

View author publications

查看作者出版物

You can also search for this author in

您也可以在中搜索此作者

PubMed

PubMed

Google Scholar

谷歌学者

Corresponding author

通讯作者

Correspondence to

通信对象

Lin Tang

林堂（音）

.

.

Rights and permissions

权限和权限

Reprints and permissions

重印和许可

About this article

关于本文

Cite this article

引用本文

Tang, L. RNA-seq coverage prediction.

Tang，L。RNA-seq覆盖率预测。

Nat Methods

Nat方法

22

22

, 225 (2025). https://doi.org/10.1038/s41592-025-02607-4

, 225 (2025).https://doi.org/10.1038/s41592-025-02607-4

Download citation

下载引文

Published

已发布

:

:

11 February 2025

2025年2月11日

Issue Date

发布日期

:

:

February 2025

2025年2月

DOI

DOI

:

:

https://doi.org/10.1038/s41592-025-02607-4

https://doi.org/10.1038/s41592-025-02607-4

Share this article

分享这篇文章

Anyone you share the following link with will be able to read this content:

与您共享以下链接的任何人都可以阅读此内容：

Get shareable link

获取可共享链接

Sorry, a shareable link is not currently available for this article.

很抱歉，本文目前没有可共享的链接。

Copy to clipboard

复制到剪贴板

Provided by the Springer Nature SharedIt content-sharing initiative

由Springer Nature SharedIt内容共享计划提供