Regulatory grammar in human promoters uncovered by MPRA-based deep learning
Lucía Barbadilla-Martínez, Noud H.M. Klaassen, Vinícius H. Franceschini-Santos, Jeremie Breda, Hatice Yücel, Miguel Hernández-Quiles, Tijs van Lieshout, Carlos G. Urzua Traslaviña, Minh Chau Luong Boi, Maryam Akbarzadeh, Celia Hermana-Garcia-Agullo, Sebastian Gregoricchio, Marcel A. De Haas, Roy Straver, Sarah Derks, Wilbert Zwart, Emile E. Voest, Lude Franke, Michiel Vermeulen, Jeroen de Ridder, Bas van Steensel
Abstract
Promoters are the core regulatory elements of all genes. Their activity ensures the correct transcription level of each individual gene, which is essential for cellular homeostasis and responses to a wide range of signals. One of the major challenges in genomics is to build computational models that accurately predict genome-wide gene expression from the sequences of regulatory elements1. Here we present promoter activity regulatory model (PARM), a cell-type-specific deep-learning model trained on specially designed massively parallel reporter assays (MPRAs) that query human promoter sequences. PARM is experimentally and computationally lightweight so that cell-type-specific and condition-specific models can be generated that reliably predict autonomous promoter activity across the genome from the DNA sequence alone. PARM can also design purely synthetic strong promoters. We leveraged PARM to systematically identify binding sites of transcription factors that probably contribute to the activity of each natural human promoter and to detect the rewiring of these regulatory interactions after various stimuli to the cells. We also uncovered and experimentally confirmed substantial positional preferences of transcription factors that differ between activating and repressive regulatory functions and a complex grammar of motif–motif interactions. Our approach provides a highly economic strategy towards a deeper understanding of the dynamic regulation of human promoters by transcription factors. PARM is a deep-learning model trained on data from massively parallel reporter assays to help predict promoter activity in different human cell types, design synthetic promoters and identify key features of regulatory promoter grammar.