Dynamic regulatory grammar of human promoters uncovered by MPRA-trained deep learning

One of the major challenges in genomics is to build computational models that accurately predict genome-wide gene expression from the sequences of regulatory elements. Promoters play a key role in gene regulation, yet their regulatory logic remains incompletely understood. Here, we present PARM, a cell-type specific deep learning model trained on specially designed massively parallel reporter assays that query human promoter sequences. PARM is computationally light-weight and reliably predicts autonomous promoter activity across the genome from DNA sequence alone, in multiple cell types. PARM can also design purely synthetic strong promoters. We leveraged PARM to systematically identify transcription factor (TF) binding sites that likely to contribute to the activity of each natural human promoter, and to detect the rewiring of these regulatory interactions upon various stimuli to the cells. We also uncovered and experimentally confirmed striking positional preferences of TFs that differ between activating and repressive regulatory functions, as well as a complex grammar of motif-motif interactions. Our approach provides a foundation towards a deeper understanding of the dynamic regulation of human promoters by TFs.

基因组学领域的核心挑战之一，是构建可从调控元件序列精准预测全基因组基因表达的计算模型。启动子在基因调控中发挥关键作用，但其调控逻辑仍未被完全阐明。本研究推出PARM——一种基于专门设计、用于探测人类启动子序列的大规模并行报告基因测定（massively parallel reporter assays）训练得到的细胞类型特异性深度学习模型。PARM计算资源占用极低，仅依靠DNA序列即可在多种细胞类型中可靠预测全基因组范围内的自主启动子活性。此外，PARM还可设计纯合成型强启动子。我们借助PARM系统性地鉴定了可能参与每个天然人类启动子活性调控的转录因子（Transcription Factor, TF）结合位点，并检测了细胞受到各类刺激时这些调控相互作用的重塑过程。我们还揭示并经实验验证，转录因子在发挥激活与抑制调控功能时具有显著不同的位置偏好性，同时还阐明了基序-基序相互作用的复杂规则。本研究为深入理解转录因子对人类启动子的动态调控奠定了坚实基础。