Exploring chromatin regulation by ChIP-Sequencing

Transcription

1 Exploring chromatin regulation by ChIP-Sequencing From datasets quality assessment, enrichment patterns identification and multi-profiles integration to the reconstitution of gene regulatory wires describing biological systems behavior Marco Antonio Mendoza IGBMC Strasbourg; France

2 Systems biology of cell fate transitions Tran re pr og ra m m in g s-dif feren tiatio n Differentiation Functional Genomics Physiological differentiation Aberrant transformation Systems Biology

3 Systems biology of retinoic acid-induced cell fate transitions Vitamin A analog All-trans retinoic acid (ATRA) ChIP-seq datasets analysis Cell differentiation Extract and compare meaningful biological information user-chosen intensity threshold. (e.g. FindPeaks, 2008) binding sites annotation enrichment confidence relative to background patterns (measured or inferred). (e.g. MACS, 2008) spatial signal enrichment dependency concept. (e.g. BayesPeak, 2009; Hpeak, 2010) scanning of the shape of enrichment patterns. A regression-based method for peak detection in proteinchromatin interaction profiles. M. A. Mendoza-Parra et al. BMC Genomics; Nov. 2013

4 Systems biology of retinoic acid-induced cell fate transitions Vitamin A analog All-trans retinoic acid (ATRA) Cell differentiation ChIP-seq datasets analysis Extract and compare meaningful biological information user-chosen intensity threshold. (e.g. FindPeaks, 2008) enrichment confidence relative to background patterns (measured or inferred). (e.g. MACS, 2008) Genomic regions selection Linear regression fitting spatial signal enrichment dependency concept. (e.g. BayesPeak, 2009; Hpeak, 2010) scanning of the shape of enrichment patterns. A regression-based method for peak detection in proteinchromatin interaction profiles. M. A. Mendoza-Parra et al. BMC Genomics; Nov. 2013

5 Systems biology of retinoic acid-induced cell fate transitions Vitamin A analog All-trans retinoic acid (ATRA) Cell differentiation ChIP-seq datasets analysis Extract and compare meaningful biological information Xu H. et al; Bioinformatics 2008 Anders S. et al; Genome Biology 2010 Shao Z. et al; Genome Biology 2012 Compare binding events in a genomic position but also in a read-counts intensity context Robinson M. et al; Bioinformatics 2010 Nair NU. et al; PLOS One 2012 inter-profiles normalization Comparative analysis of RNA Polymerase II ChIPseq profiles by non-linear normalisation approaches Klein HU. et al; M. A. Mendoza-Parra et al. Bioinformatics Nucl Acids Res; Nov. 2011

6 Systems biology of retinoic acid-induced cell fate transitions Vitamin A analog All-trans retinoic acid (ATRA) Cell differentiation ChIP-seq datasets analysis Extract and compare meaningful biological information M. A. Mendoza-Parra et al. Nucl. Acids Res; Nov EPIMETHEUS Compare binding events in a genomic position but also in a read-counts intensity context Concept expanded to other ChIPseq datasets (e.g. Histone modifications) Saleem M. et al; BMC Bioinformatics (accepted)

7 Systems biology of retinoic acid-induced cell fate transitions Vitamin A analog All-trans retinoic acid (ATRA) Cell differentiation ChIP-seq datasets analysis Extract and compare meaningful biological information QC? evaluate the quality of compared datasets which may impact the assessment of the relevant biological information a bioinformatics-based quality control (QC) system that uses raw NGS dataset information to have quantitative means for comparing different ChIP-seq profiles. A quality control system for profiles obtained by ChIP sequencing M. A. Mendoza-Parra et al. Nucl. Acids Res; Nov. 2013

8 Are you concerned about these differences in ChIP-seq assays performance? Profiles of three publicly available H3K4me3 ChIP-seq datasets. Read-count intensity ,950k 37,050k 37,150k Human genome; hg19;chr19 dataset 1 dataset 2 dataset 3 dataset 4 (WCE) Common binding events are retrieved, However there are significant differences : Global read-count intensity Observed background levels With the same commercial antibody source (including batch reference). Produced from a similar number of mapped reads (~19 million). Performed in different laboratories.

9 Source of technical variability in immuno-selection-nsg based approaches ChIP-seq workflow Protein-chromatin crosslinking Chromatin fragmentation immunoprecipitation ChIP-seq grade Abs? DNA purification Parallel sequencing Computational analysis Quality ChIP-seq profile=f( chrom shearing + cell fixation + user skills + Ab efficacy + Seq depth + ) Ab against H3K4me3 Batch A Batch B Source: Diagenode H3K4me3 antibody; technical data sheet

10 Source of technical variability in immuno-selection-nsg based approaches ChIP-seq workflow Sequencing depth? Protein-chromatin crosslinking Chromatin fragmentation immunoprecipitation DNA purification Parallel sequencing Computational analysis Quality ChIP-seq profile=f( chrom shearing + cell fixation + user skills + Ab efficacy + Seq depth + ) number of identified peaks increases with the sequencing depth Peaks confidence is proportional to the sequencing depth Rozowsky J. et al. Nature Biotechnol 2009 Zhang Y. et al. Genome Biol 2008

11 Source of technical variability in immuno-selection-nsg based approaches ChIP-seq workflow Sequencing depth? Protein-chromatin crosslinking Chromatin fragmentation immunoprecipitation ChIP-seq grade Abs? DNA purification Parallel sequencing Computational analysis Quality ChIP-seq profile=f( chrom shearing + cell fixation + user skills + Ab efficacy + Seq depth + ) as individual profiles could present significant technical variability (due to the use of different antibodies, sequencing depth or other factors able to influence the immunoprecipitation (IP) quality), comparative analyses between Next generation sequencing (NGS) generated profiles may require prior characterization of the degree of technical similarity of the various data sets

12 Current practices for the evaluation of the quality of ChIP-seq datasets: Peak calling-based approaches Visual inspection global ChIP enrichment evaluated by the Fraction of mapped reads in identified Peaks (FRiP) Drawback: only in a local context not quantitative Quality Control systems The ENCODE consortium recommends a 1% FRiP threshold Evaluate the binding sites reproductibility between biological replicate datasets (IDR) IDR: Irreproductibility Discovery Rate RAD21 replicates SPT20 replicates Drawback: they depend on the peak calling performance thus their quality assessement may not represent a universal approach for Comparing any type of ChIP-seq datasets high reproductibility low reproductibility Landt S. et al. Genome Research 2012

13 A quality control (QC) system for the analysis and comparison of NGS-generated profiles Aim of the methodology: Evaluate changes in enrichment patterns when a fraction of the total mapped reads (TMRs) are used for its reconstruction. Sequenced reads random sampling 90% 70% 50% In an ideal case the reconstructed profile will be reproduced after random sampling with a proportional decrease in the read-count intensity Read-counts intensity the deviation from the ideal behavior reflects the degree of reproducibility/robustness of the profile under study reconstructed profile s100 s90 s70 s50 It provides a quantitative method for assigning quality descriptors

14 A quality control (QC) system for the analysis and comparison of NGS-generated profiles Computational treatment: Segmented read-count intensity profile (500nts window size) RCI: read count intensity S100: Original dataset S90: 90% reads sampled S70: 70% reads sampled S50: 50% reads sampled Analysis on ChIP-seq segments (500bp bins) Read counts Int. after sampling (%) Changes in RCI after random sampling s90 s70 s S100 read counts intensity (log2) 90% 70% 50% Dispersion (%) RCI dispersion relative to ideal situation s90 s70 s s100 read counts intensity (log2) Fraction of genomic windows in a given dispersion interval

15 A universal NGS-QCi system for the stratification of current and past generated profiles H4K20me1 H3K4me2 H3K27me3 H4K5ac RXRa AR RNA PolII H3K9ac H3K4me1 H3K27ac CTCF RARg p300 RNA-seq H3K4me3 H3K36me3 H2A.Z ERa FoxA1 Med12 GRO-seq wce Global Quality control indicator Global Quality control indicator the QCi analysis provides important quality information about datasets for the same target at different sequencing depths TMRs (millions)

16 A universal NGS-QCi system for the stratification of current and past generated profiles H4K20me1 H3K4me2 H3K27me3 H4K5ac RXRa AR RNA PolII H3K9ac H3K4me1 H3K27ac CTCF RARg p300 RNA-seq H3K4me3 H3K36me3 H2A.Z ERa FoxA1 Med12 GRO-seq wce Global Quality control indicator Global Quality control indicator fequency Ab-independent datasets (e.g. RNA-seq) A B C D No enrichment control datasets (whole cell extract: wce) 1st quartile 2nd quartile 3th quartile 4th quartile Discretized global QCis (QCi-STAMP)

17 A universal NGS-QCi system for the stratification of current and past generated profiles Global Quality control indicator QCis assessed at three different confidence threshold conditions

18 A universal NGS-QCi database for comparing current and past generated profile s indicators NGS-datasets repositories Automated certification pipeline Datasets Quality? NGS-QCi database

19 (currently hosting > 57,000 entries)

20 NGS-QC Generator: A quality control system for ChIP-seq and NGS-related datasets Mus musculus (42%) Homo sapiens (44%) quality grades for more than 57,000 publicly available datasets (May 2017) More than 85% of all ChIP-seq assays available in the public repository GEO has been qualified!

21 A web-browser access for running the NGS-QC Generator on your favourite NGS dataset!!!

22 Expanding the NGS-QC Generator concept to long-range chromatin interaction assays Long-range genome interactions quality assessment Database hosting QC scores for > 250 Hi-C, >50 ChIA-PET and >640 4Cseq assays. (web-access visualization module) Mendoza-Parra et al.; BMC Genomics 2016

23 Using the NGS-QC certification system for validating ChIP-seq profiles prior further analysis Omics datasets integration Compared profiles present at least a «BBB» certification!!!!

24 Using the NGS-QC certification system for validating ChIP-seq profiles prior further analysis Omics datasets integration Combinatorial Chromatin state analysis Taudt A. et al; Nature Genetics review; 2016 Interpreting chromatin structure and function from a combinatorial analysis of ChIP-seq readouts ChIP-seq readout ChromHMM (Ernst J. et al; Nat Biotech 2010; Nat.Methods 2012) hihmm (Sohn KA et al; Bioinformatics 2015) jmosaics (Zeng X et al; Genome Biol; 2013) Segway ( Hoffman MM et al; Nat. Combinatorial chromatin statesmethods; 2012) TreeHMM ( Biesinger J. et al; BMC Bioinformatics; 2013)

25 Using the NGS-QC certification system for validating ChIP-seq profiles prior further analysis Omics datasets integration Combinatorial Chromatin state analysis Combinatorial state functional state Mendoza Parra et al; Genome Res; 2016

26 Public repositories are hosting several thousands of functional genomics datasets Assessing functional chromatin states over publicly available datasets Public access Datasets repositories ChIP-seq and related datasets released in the public domain

27 Difficulties for exploring the content of Public datasets repositories a single raw dataset covers several Gbs. Public access Datasets repositories when processed data are available, multiprofile comparisons might suffer from variable (unkown) datasets preprocessing conditions. no information about its quality is available. not all of us have the computational skills neither the required computational resources for processing multiple datasets recovered from the public domain.

28 Providing means for exploring the content of publicly available NGS-datasets repositories Datasets qualification (automated pipeline) An intuitive solution for retrieving public datasets on the basis of: Target molecule Cell-type/tissue Repository ID quality grade (AAA CCC?) author keywords in abstract Datasets Exploration H3K27me3 H3K4me3 H3K27ac mm9 Hoxa1

29 (automated pipeline) Providing means for exploring the content of publicly available NGS-datasets repositories Datasets Exploration H3K27me3 H3K4me3 H3K27ac mm9 Hoxa1 Enrichment representation (localqc) Local Quality high low

30 Providing means for exploring the content of publicly available NGS-datasets repositories Datasets Exploration ChIP-seq profiles pair-wise similarity evaluation

31 Providing means for exploring the content of publicly available NGS-datasets repositories Datasets Integration Assessing functional chromatin states over several (up to hundreds) public datatsets All these analyses can be performed directly on a web-browser access. No need to download the required profiles; just use the qc-explorer interphase to retrieve the profiles of interest.

32 Providing means to the scientific community for exploring and comparing several hundreds of datasets retrieved on public repositories Data integration itself is not an end: it is designed to generate novel hypotheses and help to test them. If a hypothetical Data Integrator existed, its most important input would not be the data to be analyzed, but a specific question to answer D. Hawkins, Gary C. Hon and Bing Ren Nat. Rev. Genet. July 2010

33 acknowledgements Benjamin Billore Wouter vangool Ashick Saleem Martial Sankar Malgorzata Nowicka Michèle Lieb Functional Genomics & Cancer Spin-off Mathias Blum Pierre-etienne Cholley Valeriya Malysheva

34