The dynseq browser track shows context-specific features at nucleotide resolution

Abstract

High-throughput experimental platforms have revolutionized the ability to profile biochemical and functional properties of biological sequences such as DNA, RNA and proteins. By collating several data modalities with customizable tracks rendered using intuitive visualizations, genome browsers enable an interactive and interpretable exploration of diverse types of genome profiling experiments and derived annotations. However, existing genome browser tracks are not well suited for intuitive visualization of high-resolution DNA sequence features such as transcription factor motifs. Typically, motif instances in regulatory DNA sequences are visualized as BED-based annotation tracks, which highlight the genomic coordinates of the motif instances but do not expose their specific sequences. Instead, a genome sequence track needs to be cross-referenced with the BED track to identify sequences of motif hits. Even so, quantitative information about the motif instances such as affinity or conservation as well as differences in base resolution from the consensus motif are not immediately apparent. This makes interpretation slow and challenging. This problem is compounded when analyzing several cellular states and/or molecular readouts (such as ATAC-seq and ChIP–seq) simultaneously, as coordinates of enriched regions (peaks) and the set of active transcription factor motifs vary across cell states.

Recently, machine learning models that map DNA sequence to functional readouts from high-throughput assays have been developed to study the sequence basis of molecular activity and decipher putative functional genetic variants that influence protein–DNA binding, splicing, gene expression and long-range chromatin contacts^1-3. These models are interrogated using feature attribution methods to infer quantitative, predictive importance scores of each base, thereby enabling the discovery of sequence features such as transcription factor motifs, splice sites and polyadenylation sites^4,5. These importance scores are currently visualized in ad hoc ways that are not suited to seamless exploration and easy sharing.

To address these challenges, we introduce the dynamic sequence (dynseq) genome browser track, a generalization of previously proposed ‘sequence walkers’⁶ adapted to modern genome browsers, that displays DNA nucleotide characters at a genomic locus with heights scaled by user-specified, base-resolution, quantitative scores. The dynseq track makes it straightforward to visually recognize sequence features that are activated in a context-specific manner.

To visualize context-specific dynamic importance scores, we first implemented and integrated the dynseq track in the WashU Epigenome Browser^7,8. The input file format is the BigWig format⁹ with per-base-pair importance scores. At each position, the nucleotide character (A/C/G/T) is rendered. The height of the character is scaled by the importance score at that position. Negative scores are handled by flipping the character along the x axis. When the track is zoomed out such that individual bases cannot be discerned, the track visualization switches to a regular BigWig view. This simple specification means that the dynseq track can be easily incorporated into other genome browsers. We have added equivalent native functionality for the UCSC Genome Browser¹⁰ and HiGlass¹¹.

To illustrate a typical use case, we trained separate BPNet neural networks¹² to map DNA sequence to base-resolution profiles of DNase sequencing (DNase-seq) and five transcription factor chromatin immunoprecipitation with high-throughput sequencing (ChIP–seq) experiments in the K562 cell line^13,14. We used the DeepLIFT feature attribution algorithm^15,16 to derive nucleotide importance scores for sequences underlying regions with enriched signal (peaks) from each of the models. We used the WashU Epigenome Browser to visualize and interpret an enhancer in the β-globin locus approximately 10 kb upstream of the HBE1 gene¹⁷. We used BigWig tracks to display the observed and model-predicted profiles for each assay, and dynseq tracks to visualize assay-specific importance scores derived from each model (Fig. 1 and Supplementary Fig. 1). The dynseq tracks highlight the predicted quantitative influence of different motifs on chromatin accessibility and transcription factor binding profiles. The binding profiles of NFE2, GATA1, GATA2 and USF1 are predicted to be influenced by their respective direct binding motifs. By contrast, TAL1 is affected by GATA and several weaker TAL1 motifs, which suggests that GATA transcription factors are potentially involved in cooperatively recruiting TAL1 to their binding sites. The dynseq track of the DNase-seq model highlights GATA and NFE motifs, which suggests that these motifs primarily drive DNase I hypersensitivity at this locus. We also visualized DNase-seq footprinting scores¹⁸ and PhyloP conservation scores¹⁹ using both the standard BigWig rendering and the dynseq track. The footprinting scores highlight NFE motifs, whereas the PhyloP dynseq track highlights strong conservation of GATA, NFE and several TAL motifs. Together, dynseq tracks reveal subtle, quantitative, context-specific sequence determinants of transcription factor binding and chromatin accessibility, thereby providing insights into the architecture of cis-regulatory elements. Its generality allows for versatile applications, including the exploration of sequence features disrupted by variants that influence regulatory activity via allele-aware dynseq tracks in the Resgen/FliGlass browser²⁰ (Supplementary Note, Supplementary Fig. 2).

Fig. 1 ∣. WashU Epigenome Browser session for deciphering sequence architecture of a cis-regulatory element.

a, Human β-globin locus with observed and predicted DNase-seq tracks and zoomed out DNase-seq model-derived importance dynseq track. b, Enhancer 10 kb upstream of HBE1 (hg38 chr11:5280607–5280830) with observed base-resolution 5′ end coverage tracks of DNase-seq and ChIP–seq targeting transcription factors NFE2, GATA1, GATA2, TAL1 and USF1 in the K562 cell line; corresponding predicted tracks from BPNet sequence models trained on these data; BPNet model-derived nucleotide importance scores visualized using dynseq tracks; and DNase-seq footprinting scores, −log(P value) and PhyloP conservation scores visualized using standard BigWig and dynseq tracks. For ChIP–seq profile tracks, blue denotes plus (+) and orange denotes minus (−) strand.

Integration of dynseq tracks into genome browsers enables intuitive, sequence-centric visualization of diverse quantitative, base-resolution scores, including sequence contribution scores from predictive models, sequence conservation scores, motif match scores and high-resolution transcription factor footprinting. We expect that dynseq tracks will enhance exploratory analysis, discovery and hypothesis generation by enabling contextual interpretation of informative sequence features in genomic elements and those disrupted by variation at single-nucleotide resolution.

Data availability

Data and models used to create vignettes are available at: https://doi.org/10.5281/zenodo.6582100. See Supplementary Table 1 for browser-specific functionalities

Code availability

Code to reproduce vignettes is available at: https://doi.org/10.5281/zenodo.7019993. The dynseq tracks use the BigWig file format. A tutorial is available at: https://kundajelab.github.io/dynseq-pages/. The dynseq track is supported by:

• UCSC Genome Browser (https://genome.ucsc.edu). Documentation is available at: https://genome.ucsc.edu/goldenpath/help/big-Wig.html#dynseq. Source code is available at: https://github.com/ucscGenomeBrowser/kent.

• HiGlass/Resgen (https://higlass.io; https://resgen.io). Dynseq is implemented as a plugin. Source code is available at: https://github.com/kundajelab/higlass-dynseq/.

• WashU Epigenome Browser (https://epigenomegateway.wustl.edu). Source code is available at: https://github.com/lidaof/eg-react. Documentation is available at: https://eg.readthedocs.io/en/latest/tracks.html#dynseq.