iCREPCP is a deep learning-based web server which has learned the cis-regulatory code of plant core promoters from three plant species of Arabidopsis, maize and sorghum. As an application, it can accurately identify which cis-regulatory elements (CREs) a given plant core promoter contains, as well as locate the exact position of each CRE (at base resolution) and its contribution to the plant core promoter's activity. Our website is www.hzau-hulab.com/icrepcp/.
Citation: Kaixuan Deng#, Qizhe Zhang#, Yuxin Hong, Jianbing Yan* and Xuehai Hu*(2022), iCREPCP: a deep learning-based web server for identifying critical cis-regulatory elements within plant core promoters.
- tensorflow==2.4 (for model training)
- tensorflow==1.14, keras==2.3.1, deeplift==0.6.13.0 (for deepLIFT and TF-moDISco)
- pandas
- matplotlib
- seaborn
git clone git@https://github.com/kaixuanDeng95/iCREPCP.git
Cis-regulatory modules(CRMs) and trans-acting factors(TAFs) play an important role in specifying the quantitative level of gene expressions in plant biology.Common CRMs include gene-proximal promoters, distal enhancers, and silencers, which are all considered as the complex assemblies of cis-regulatory elements (CREs). The identification and characterization of plant CRMs or critical CREs will not only provide helpful knowledge for our understandings of transcriptional regulatory mechanisms in plants, but also is an essential prerequisite for plant breeding by genome editing.
Our focus is on Plant Core Promoters (PCP), a large group of CRMs, which can drive basal level of target gene transcriptions because they are rich in CREs. A typical PCP is the minimal sequence region (usually 50-100bp around transcription start site (TSS) of plant genes) needed to direct initiation of transcription. The promoter strength of PCP is defined as the ability to drive expression of a barcoded green fluorescent protein (GFP) reporter gene in tobacco leaves or maize protoplasts systems.
Here, we first trained a deep learning model to learn the cis-regulatory code of PCP and then developed a deep learning-based web server to identify which CREs a given Plant Core Promoter (iCREPCP) contains, with a focus on the exact position of each CRE and its contribution to the promoter strength. To achieve the above purpose, we first employed a large-scale dataset of 18,329 Arabidopsis, 34,415 maize and 27,094 sorghum core promoters, whose strengths were measured by STARR-seq assays via two transient transfection systems of tobacco leaves and maize protoplasts. We then trained a deep learning model of ‘DenseNet’ to accurately fit promoter strengths with their DNA sequences. Importantly, a powerful model interpretability tool of ‘DeepLIFT’ was employed to assign feature importance (via base contribution score) to each base in view of its contribution to make prediction of the promoter strength with the trained model. Finally, several successive bases with significantly high values of base contribution score were identified as critical CREs within a given PCP by ‘TF-MoDISco’ (Transcription Factor Motif Discovery from Importance Scores).
If you have any questions or suggestions, please mail to 1695756050@qq.com or huxuehai@mail.hzau.edu.cn.