Dissection of core promoter syntax through single nucleotide resolution modeling of transcription initiation

Abstract

Our understanding of how the DNA sequences of cis-regulatory elements encode transcription initiation patterns remains limited. Here we introduce CLIPNET, a deep learning model trained on population-scale PRO-cap data that accurately predicts the position and quantity of transcription initiation with single nucleotide resolution from DNA sequence. Interpretation of CLIPNET revealed a complex regulatory syntax consisting of DNA-protein interactions in five major positions between -200 and +50 bp relative to the transcription start site, as well as more subtle positional preferences among different transcriptional activators. Transcriptional activator and core promoter motifs occupy different positions and play distinct roles in regulating initiation, with the former driving initiation quantity and the latter initiation position. We identified core promoter motifs that explain initiation patterns in the majority of promoters and enhancers, including DPR motifs and AT-rich TBP binding sequences in TATA-less promoters. Our results provide insights into the sequence architecture governing transcription initiation.

Deep-learning prediction of gene expression from personal genomes

“…For each gene, we fetched an individual's two 49-kilobase (kb) consensus sequences centered on the gene's TSS (GENCODE 30 v26). We one-hot encoded each sequence, and used the average of two one-hot encoded matrices as our input 24 . While other approaches are possible 7 , we found this representation to be reasonable because:…”

“…Previous work used far fewer individuals and did not evaluate across them. 7,24 To address this we developed Performer, a fine-tuning strategy that implements cross-individual training and evaluation of sequence-to-expression neural network models. Briefly, we modified the Enformer architecture 15 by replacing the output head with one that predicts tissue-specific gene expression as a scalar value rather than a genomic track and implemented fine-tuning with Enformer's weights as starting values for the parameters in the model trunk and a custom loss function (Methods).…”