Differential ATAC-seq and ChIP-seq peak detection using ROTS

Changes in cellular chromatin states fine-tune transcriptional output and ultimately lead to phenotypic changes. Here we propose a novel application of our reproducibility-optimized test statistics (ROTS) to detect differential chromatin states (ATAC-seq) or differential chromatin modification states (ChIP-seq) between conditions. We compare the performance of ROTS to existing and widely used methods for ATAC-seq and ChIP-seq data using both synthetic and real datasets. Our results show that ROTS outperformed other commonly used methods when analyzing ATAC-seq data. ROTS also displayed the most accurate detection of small differences when modeling with synthetic data. We observed that two-step methods that require the use of a separate peak caller often more accurately called enrichment borders, whereas one-step methods without a separate peak calling step were more versatile in calling sub-peaks. The top ranked differential regions detected by the methods had marked correlation with transcriptional differences of the closest genes. Overall, our study provides evidence that ROTS is a useful addition to the available differential peak detection methods to study chromatin and performs especially well when applied to study differential chromatin states in ATAC-seq data.

ChIP-AP – An Integrated ChIP-Seq Analysis Pipeline

ChIP-Seq is a technique used to analyse protein-DNA interactions. The protein-DNA complex is pulled down using a protein antibody, after which sequencing and analysis of the bound DNA fragments is performed. A key bioinformatics analysis step is “peak” calling - identifying regions of enrichment. Benchmarking studies have consistently shown that no optimal peak caller exists. Peak callers have distinct selectivity and specificity characteristics which are often not additive and seldom completely overlap in many scenarios. In the absence of a universal peak caller, we rationalized one ought to utilize multiple peak-callers to 1) gauge peak confidence as determined through detection by multiple algorithms, and 2) more thoroughly survey the protein-bound landscape by capturing peaks not detected by individual peak callers owing to algorithmic limitations and biases. We therefore developed an integrated ChIP-Seq Analysis Pipeline (ChIP-AP) which performs all analysis steps from raw fastq files to final result, and utilizes four commonly used peak callers to more thoroughly and comprehensively analyse datasets. Results are integrated and presented in a single file enabling users to apply selectivity and sensitivity thresholds to select the consensus peak set, the union peak set, or any sub-set in-between to more confidently and comprehensively explore the protein-bound landscape. (https://github.com/JSuryatenggara/ChIP-AP).

WACS: improving ChIP-seq peak calling by optimally weighting controls

Background Chromatin immunoprecipitation followed by high throughput sequencing (ChIP-seq), initially introduced more than a decade ago, is widely used by the scientific community to detect protein/DNA binding and histone modifications across the genome. Every experiment is prone to noise and bias, and ChIP-seq experiments are no exception. To alleviate bias, the incorporation of control datasets in ChIP-seq analysis is an essential step. The controls are used to account for the background signal, while the remainder of the ChIP-seq signal captures true binding or histone modification. However, a recurrent issue is different types of bias in different ChIP-seq experiments. Depending on which controls are used, different aspects of ChIP-seq bias are better or worse accounted for, and peak calling can produce different results for the same ChIP-seq experiment. Consequently, generating “smart” controls, which model the non-signal effect for a specific ChIP-seq experiment, could enhance contrast and increase the reliability and reproducibility of the results. Result We propose a peak calling algorithm, Weighted Analysis of ChIP-seq (WACS), which is an extension of the well-known peak caller MACS2. There are two main steps in WACS: First, weights are estimated for each control using non-negative least squares regression. The goal is to customize controls to model the noise distribution for each ChIP-seq experiment. This is then followed by peak calling. We demonstrate that WACS significantly outperforms MACS2 and AIControl, another recent algorithm for generating smart controls, in the detection of enriched regions along the genome, in terms of motif enrichment and reproducibility analyses. Conclusions This ultimately improves our understanding of ChIP-seq controls and their biases, and shows that WACS results in a better approximation of the noise distribution in controls.

LanceOtron: a deep learning peak caller for ATAC-seq, ChIP-seq, and DNase-seq

Genomics technologies, such as ATAC-seq, ChIP-seq, and DNase-seq, have revolutionized molecular biology, generating a complete genome’s worth of signal in a single assay. Coupled with the use of genome browsers, researchers can now see and identify important DNA encoded elements as peaks in an analog signal. Despite the ease with which humans can visually identify peaks, converting these signals into meaningful genome-wide peak calls from such massive datasets requires complex analytical techniques. Current methods use statistical frameworks to identify peaks as sites of significant signal enrichment, discounting that the analog data do not follow any archetypal distribution. Recent advances in artificial intelligence have shown great promise in image recognition, on par or exceeding human ability, providing an opportunity to reimagine and improve peak calling. We present an interactive and intuitive peak calling framework, LanceOtron, built around image recognition using a wide and deep neural network. We hand-labelled 499Mb of genomic data, built 5,000 models, and tested with over 100 unique users from labs around the world. In benchmarking open chromatin, transcription factor binding, and chromatin modification datasets, LanceOtron outperforms the long-standing, gold-standard peak caller MACS2 with its increased selectivity and near perfect sensitivity. Additionally, this command-line optional approach allows researchers to easily generate optimal peak-calls using only a web interface. Together, the enhanced performance, and usability of LanceOtron will improve the reliability and reproducibility of peak calls and subsequent data analysis. This tool highlights the general utility of applying machine learning to genomic data extraction and analysis.

F-Seq2: improving the feature density based peak caller with dynamic statistics

Genomic and epigenomic features are captured at a genome-wide level by using high-throughput sequencing (HTS) technologies. Peak calling delineates features identified in HTS experiments, such as open chromatin regions and transcription factor binding sites, by comparing the observed read distributions to a random expectation. Since its introduction, F-Seq has been widely used and shown to be the most sensitive and accurate peak caller for DNase I hypersensitive site (DNase-seq) data. However, the first release (F-Seq1) has two key limitations: lack of support for user-input control datasets, and poor test statistic reporting. These constrain its ability to capture systematic and experimental biases inherent to the background distributions in peak prediction, and to subsequently rank predicted peaks by confidence. To address these limitations, we present F-Seq2, which combines kernel density estimation and a dynamic ‘continuous’ Poisson test to account for local biases and accurately rank candidate peaks. The output of F-Seq2 is suitable for irreproducible discovery rate analysis as test statistics are calculated for individual candidate summits, allowing direct comparison of predictions across replicates. These improvements significantly boost the performance of F-Seq2 for ATAC-seq and ChIP-seq datasets, outperforming competing peak callers used by the ENCODE Consortium in terms of precision and recall.

F-Seq2: improving the feature density based peak caller with dynamic statistics

ABSTRACTGenomic and epigenomic features are captured at a genome-wide level by using high-throughput sequencing technologies. Peak calling is one of the first essential steps in analyzing these features by delineating regions such as open chromatin regions and transcription factor binding sites. Our original peak calling software, F-Seq, has been widely used and shown to be the most sensitive and accurate peak caller for DNase I hypersensitive sites sequencing (DNase-seq) data. However, F-Seq lacks support for user-input control dataset nor reporting test statistics, limiting its ability to capture systematic and experimental biases and accurately estimate background distributions. Here we present an improved version, F-Seq2, which combined the power of kernel density estimation and a dynamic “continuous” Poisson distribution to robustly account for local biases and solve ties when ranking candidate peaks. In F-score and motif distance analysis, we demonstrated the superior performance of F-Seq2 than other competing peak callers used by the ENCODE Consortium on simulated and real ATAC-seq and ChIP-seq datasets. The output of F-Seq2 is suitable for irreproducible discovery rate (IDR) analysis as the test statistics calculated for individual candidate summit and ties are robustly solved.

Comparative analysis of ChIP-exo peak-callers: impact of data quality, read duplication and binding subtypes

Background: ChIP (Chromatin immunoprecipitation)-exo has emerged as an important and versatile improvement over conventional ChIP-seq as it reduces the level of noise, maps the transcription factor (TF) binding location in a very precise manner, upto single base-pair resolution, and enables binding mode prediction. Availability of numerous peak-callers for analyzing ChIP-exo reads has motivated the need to assess their performance and report which tool executes reasonably well for the task. Results: This study has focussed on comparing peak-callers that report direct binding events with those that report indirect binding events. The effect of strandedness of reads and duplication of data on the performance of peak-callers has been investigated. The number of peaks reported by each peak-caller is compared followed by a comparison of the annotated motifs present in the reported peaks. The significance of peaks is assessed based on the presence of a motif in top peaks. Indirect binding tools have been compared on the basis of their ability to identify annotated motifs and predict mode of protein-DNA interaction. Conclusion: By studying the output of the peak-callers investigated in this study, it is concluded that the tools that use self-learning algorithms, i.e. the tools that estimate all the essential parameters from the aligned reads, perform better than the algorithms which require formation of peak-pairs. The latest tools that account for indirect binding of TFs appear to be an upgrade over the available tools, as they are able to reveal valuable information about the mode of binding in addition to direct binding. Furthermore, the quality of ChIP-exo reads have important consequences on the output of data analysis.

Comparative analysis of ChIP-exo peak-callers: impact of data quality, read duplication and binding subtypes

Background: ChIP (Chromatin immunoprecipitation)-exo has emerged as an important and versatile improvement over conventional ChIP-seq as it reduces the level of noise, maps the transcription factor (TF) binding location in a very precise manner, upto single base-pair resolution, and enables binding mode prediction. Availability of numerous peak-callers for analyzing ChIP-exo reads has motivated the need to assess their performance and report which tool executes reasonably well for the task. Results: This study has focussed on comparing peak-callers that report direct binding events with those that report indirect binding events. The effect of strandedness of reads and duplication of data on the performance of peak-callers has been investigated. The number of peaks reported by each peak-caller is compared followed by a comparison of the annotated motifs present in the reported peaks. The significance of peaks is assessed based on the presence of a motif in top peaks. Indirect binding tools have been compared on the basis of their ability to identify annotated motifs and predict mode of protein-DNA interaction. Conclusion: By studying the output of the peak-callers investigated in this study, it is concluded that the tools that use self-learning algorithms, i.e. the tools that estimate all the essential parameters from the aligned reads, perform better than the algorithms which require formation of peak-pairs. The latest tools that account for indirect binding of TFs appear to be an upgrade over the available tools, as they are able to reveal valuable information about the mode of binding in addition to direct binding. Furthermore, the quality of ChIP-exo reads have important consequences on the output of data analysis.

Comparative analysis of ChIP-exo peak-callers: impact of data quality, read duplication and binding subtypes

Background: ChIP (Chromatin immunoprecipitation)-exo has emerged as an important and versatile improvement over conventional ChIP-seq as it reduces the level of noise, maps the transcription factor (TF) binding location in a very precise manner, upto single base-pair resolution, and enables binding mode prediction. Availability of numerous peak-callers for analyzing ChIP-exo reads has motivated the need to assess their performance and report which tool executes reasonably well for the task. Results: This study has focussed on comparing peak-callers that report direct binding events with those that report indirect binding events. The effect of strandedness of reads and duplication of data on the performance of peak-callers has been investigated. The number of peaks reported by each peak-caller is compared followed by a comparison of the annotated motifs present in the reported peaks. The significance of peaks is assessed based on the presence of a motif in top peaks. Indirect binding tools have been compared on the basis of their ability to identify annotated motifs and predict mode of protein-DNA interaction. Conclusion: By studying the output of the peak-callers investigated in this study, it is concluded that the tools that use self-learning algorithms, i.e. the tools that estimate all the essential parameters from the aligned reads, perform better than the algorithms which require formation of peak-pairs. The latest tools that account for indirect binding of TFs appear to be an upgrade over the available tools, as they are able to reveal valuable information about the mode of binding in addition to direct binding. Furthermore, the quality of ChIP-exo reads have important consequences on the output of data analysis.