Universal correction of enzymatic sequence bias reveals molecular signatures of protein/DNA interactions

Abstract

Coupling molecular biology to high throughput sequencing has revolutionized the study of biology. Molecular genomics techniques are continually refined to provide higher resolution mapping of nucleic acid interactions and structure. Sequence preferences of enzymes can interfere with the accurate interpretation of these data. We developed seqOutBias to characterize enzymatic sequence bias from experimental data and scale individual sequence reads to correct intrinsic enzymatic sequence biases. SeqOutBias efficiently corrects DNase-seq, TACh-seq, ATAC-seq, MNase-seq, and PRO-seq data. We show that seqOutBias correction facilitates identification of true molecular signatures resulting from transcription factors and RNA polymerase interacting with DNA. IntroductionThe field of molecular genomics emerged as classical molecular biology techniques were coupled to high throughput sequencing technology to provide unprecedented genome-wide measurements of molecular features. Molecular genomics assays, such as DNase-seq (1, 2), ChIP-exo (3), and PRO-seq (4, 5), are converging on single-nucleotide resolution measurements. The enzymes that are routinely used in molecular biology and cloning have inherent and often uncharacterized sequence preferences. These preferences manifest more prominently as the resolution of genomic assays increases. Therefore, we developed seqOutBias (https://github.com/guertinlab/seqOutBias) to characterize and correct enzymatic biases that can obscure proper interpretation of molecular genomics data.Enzymatic hypersensitivity assays, such as DNase-seq (1, 2), TACh-seq (6), and ATAC-seq (7), have the potential to measure transcription factor (TF) binding sites genome-wide in a single experiment. These assays strictly measure enzymatic (DNase, Tn5 transposase, Benzonase, or Cyanase) accessibility to DNA and not a specific biological event, making data challenging to deconvolve. Standard algorithms scan for footprints, which are depletions of signal in larger regions of hypersensitivity (8-12). Many transcription factors, however, do not exhibit composite footprints if enzymatic cut frequency is averaged at all ChIP-seq validated binding sites with strong consensus motifs (10-13). Moreover, the inability to detect a footprint at any individual TF binding site results in high false negative rates for footprinting algorithms (14). Accurate footprinting is also confounded by the artifactual molecular signatures that result from enzymatic sequence preference (10-12). DNase footprinting algorithms can incorporate DNase cut preference data to abrogate this bias (12, 15). However, no existing tools specialize in correcting intrinsic sequence bias for a diverse set of enzymes and experimental methodologies.We find that correcting for enzymatic sequence bias highlights true molecular signatures that result from TF/DNA interactions. Despite the limitations of enzymatic hypersensitivity footprinting and sequence bias signatures, hypersensitive regions reveal a near-comprehensive set of functional regulatory regio...

HMMRATAC: a Hidden Markov ModeleR for ATAC-seq

“…The Tn5 transposase also has a sequence bias, which would need to be corrected before an effective footprint analysis could occur. Several methods exist(45,46) that could correct the sequence bias from an ATAC-seq library. HMMRATAC could be used downstream of such sequence corrections, to reduce bias’ in peak calling, or upstream of the corrections, to identify the reads within peaks that need to be corrected for footprint or other downstream analysis.…”

Defining data-driven primary transcript annotations with primaryTranscriptAnnotation in R

“…We strand-separated the aligned bam files using . We used (Martins et al, 2018) to simultaneously shift the alignments to represent the 3′end of the RNA and convert the BAM to the bigWig format. We merged the strand-specific bigWig files from all replicates using the UCSC Genome Browser Utilities (Kent et al, 2010).…”