Deciphering the Sequence Basis and Application of Transcriptional Initiation Regulation in Plant Genomes Through Deep Learning

Transcription initiation is a critical regulatory step in plant gene expression, yet its sequence<br>determinants remain largely elusive. Here we introduce GenoRetriever, an interpretable deep<br>learning model that deciphers the sequence basis of transcriptional initiation regulation across<br>plant genomes. Trained on STRIPE-seq data from 16 soybean tissues and six other crop species,<br>GenoRetriever identifies 27 core sequence motifs that govern transcription start site (TSS)<br>selection and usage. The model predicts TSS locations and usage levels with high accuracy, as<br>validated by in silico motif insertions, saturation mutagenesis, and CRISPR-Cas9 promoter editing.<br>It further reveals that 31.85% of natural variation between wild and domesticated soybean drives<br>shifts in promoter motif usage during domestication, and uncovers lineage-specific motif effects<br>between monocots and dicots. This interpretable model and its user-friendly web server for<br>promoter analysis and design make GenoRetriever both a methodological innovation and practical<br>tool for plant functional genomics and crop improvement.<br>

转录起始是植物基因表达中的关键调控步骤，但其序列决定因素在很大程度上仍未明确。
本研究推出GenoRetriever，一款可解释的深度学习模型，可解析植物基因组中转录起始调控的序列基础。
该模型基于16个大豆组织及其他6种作物的STRIPE-seq数据训练而来，共识别出27个调控转录起始位点（TSS）选择与使用的核心序列基序。
该模型可高精度预测转录起始位点的位置及其使用水平，相关结论已通过计算机模拟基序插入、饱和诱变及CRISPR-Cas9启动子编辑实验得到验证。
研究进一步发现，野生大豆与栽培大豆之间31.85%的自然变异，会在驯化过程中驱动启动子基序使用模式的改变，并揭示了单子叶植物与双子叶植物间存在谱系特异性的基序效应。
这款可解释模型及其用于启动子分析与设计的用户友好型Web服务器，使GenoRetriever既成为一项方法学创新，也成为植物功能基因组学与作物遗传改良领域的实用工具。