HTSstation: a web application and open-access libraries for high-throughput sequencing data analysis

Abstract

The HTSstation analysis portal is a suite of simple web forms coupled to modular analysis pipelines for various applications of High-Throughput Sequencing including ChIP-seq, RNA-seq, 4C-seq and re-sequencing. HTSstation offers biologists the possibility to rapidly investigate their HTS data using an intuitive web application with heuristically pre-defined parameters. A number of open-source software components have been implemented and can be used to build, configure and run HTS analysis pipelines reactively. Besides, our programming framework empowers developers with the possibility to design their own workflows and integrate additional third-party software. The HTSstation web application is accessible at http://htsstation.epfl.ch.

Figure 1. HTSstation modules.

Different workflows are obtained as combinations of the two pre-processing modules (dealing with raw sequences) and the four application-specific modules (manipulating sequence alignments and genomic profiles). Results can be visualized using our genomic browser GDV or sent to external post-processing tools for downstream analysis.

Figure 2. Architecture of HTSstation.

This platform offers different access modalities. The web interface targets biologists and occasional users. Command-line execution of underlying executables (bbcfutils) will fit the needs of more advanced users. Users with programming skills can import the libraries (bbcflib) and implement new workflows. GenRep (genomic repository) provides consistency and versioning of reference genome data and Bein (workflow manager) handles dispatching, tracking and documenting programs executions and outputs in a computing environment-independent manner: analyses can be launched locally, or dispatched to a cloud or a cluster.

Figure 3. Summary of the application-specific analyses.

Each module is dedicated to the analysis of a particular type of HTS experiment. The ChIP-seq module seeks significant interactions of protein with DNA, the RNA-seq module quantifies the expression levels of transcripts and compares them between conditions, the 4C-seq module identifies physical interactions along the DNA sequence and the re-sequencing module discovers polymorphisms.

Figure 4. Design of the web interface.

Input form of the mapping module: results from a previous job are imported using its unique key (1), e.g. from a demultiplexing task. Users provide inputs as URLs and organize them hierarchically in groups and runs (2) where groups represent distinct experimental conditions (3) and runs are distinct replicates (4). When relevant, some groups can be selected as representing the control condition. Selected runs can be moved between groups (5). All modules have a section with general parameters (e.g. email and analysis name) (6) and module-specific options (7).

Figure 5. Genomic and differential analysis views of RNA-seq data.

A) Differential expression analysis of coding genes (MA-plot) shows a few significantly differential genes which is consistent with the signal displayed along the genome: 1) Alox15, on chromosome 11 is strongly over-expressed in the KO condition, unlike the next gene downstream (Pelp1) 2) Col4a1 and Col4a2 (located next to each-other on chromosome 8) are both over-expressed in the KO condition, 3) Trim28/Kap1 is the knocked-out gene and is consistently strongly suppressed. Remark that the exon structure of the genes appears clearly in the genomic view. B) Reads were also mapped and quantified on the Repbase collection of repetitive elements. Plots of read coverage along the sequence of three representative examples show that RLT1IAP is highly over-expressed in KO mice, MERVL is slightly over-expressed and SINEB1 is under-expressed.

Figure 6. Chromatin configuration and transcriptional status in the HoxD cluster.

A) Illustration of the successive steps of a 4C-seq analysis. From top to bottom: coverage profile (reads density) is smoothed by a moving average. The smoothed density is corrected by fitting a model of the local interaction profile and long-range interacting regions are identified with the domainogram algorithm. B) From top to bottom: smoothed 4C-seq profiles for Hoxd13 and Hoxd4 in AT; the ChIP-seq density profiles of H3K27me3 and H3K4me3 in the same tissues with the peaks of H3K4me3 found by MACS. This figure was generated with the GDV genome viewer.

Figure 7. Global analysis of chromatin state.

A) Heatmaps of HoxD genes segmented into 25 bins plus 5 bins of 20 bp in the gene promoters. Left panel contains the H3K4me3 average ChIP-seq values; right panel shows the same for H3K27me3. The blue bar indicates the inactive genes, the green bar the active genes. B) Scatter plot showing all genes of chromosome 2 (2 Kb segments around gene starts, the density of genes is indicated by the intensity of orange) with the HoxD genes highlighted in green (active segment) and blue (repressed segment).