Statistical Analysis Of Long-range Chromatin Interaction Data From ChIA-PET Experiments

Diverse high-throughput methods have been developed to detect genome-wide chromatin interactions,including chromatin interaction analysis by paired-end tag sequencing(ChIA-PET).ChIA-PET is an important experimental method for detecting specific proteinmediated chromatin loops genome-wide at high resolution.It can discover many chromatin interactions at a higher resolution that are needed for studying gene transcription regulation.In ChIA-PET data,there is a mixture of signals and noises.Distinguishing the true interaction pairs from the noise is not a simple task,and thus tough computational tools are needed.Therefore,we proposed a new statistical approach with a mixture model,ChIAMM,to detect significant chromatin interactions from ChIA-PET experiments data.ChIAMM used a truncated Poisson regression in a Bayesian framework to consider more systematic biases: the genomic distance,local enrichment,mappability,and GC content.Using different ChIA-PET datasets,we evaluated the performance of ChIAMM with preexisting tools(ChIA-PET Tool V3,Chia Sig,Mango,ChIA-PET2,and ChIAPo P)using CTCF coverage,CTCF motif orientation,aggregate peak analysis,and computational time.The result showed that the new approach performed better than the top existing methods in detecting significant chromatin interactions from the ChIA-PET experiment data.Mostly count data exhibits over-dispersion,and the negative binomial distribution would be better suited to fit the data than the Poisson distribution.Therefore,secondly,we proposed the updated version of ChIAMM,ChIAMM2.It used a mixture model to detect significant chromatin interactions from ChIA-PET experiments data using truncated negative binomial regression in a Bayesian framework.We considered the same systematic biases as ChIAMM,the genomic distance,local enrichment,mappability,and GC content.Using different ChIA-PET datasets,we evaluated the performance of ChIAMM2 with ChIAMM using statistical methods,aggregate peak analysis,and computational time.We found that ChIAMM2 statistically fits the ChIA-PET experiments data better and showed little difference in detecting more significant chromatin interactions.The APA plot shows very similar values between the methods.Nevertheless,ChIAMM2 needs more computational time than ChIAMM.The chromatin interactions can have different intensities in different developmental stages and types of cells as well,and these differences are associated with a disease.To understand genomic regulation,identifying the changes in chromatin interactions is the critical step.Before we measure the changes in chromatin interactions,we have to think about data normalization primarily.Therefore,thirdly,we proposed a new approach,NDID.It is used for joint normalization and differential chromatin interaction detection from ChIA-PET experiments.We applied a novel approach,ME plot that displays the interaction differences between two ChIA-PET data on a single plot.The ME plot used the locally weighted linear regression(loess)to consider the anchor enrichment bias.We tested the performance of NDID on different ChIA-PET datasets,and the ME plot showed a straight loess regression curve and centered cloud point around the line.We computed the CTCF coverage analysis and found more differential interaction anchors covered with the CTCF peaks.We found that the differential regions contain genes involved in the different systems,biological and molecular processes from the pathway enrichment and GO analysis.Therefore,these results assured that the NDID could normalize the joint ChIA-PET datasets and remove the data-driven bias successfully.Generally,in this study,we established the three novels statistical methods,ChIAMM,ChIAMM2,and NDID,for ChIA-PET experiment data analysis through applying the most advanced statistical methods.These methods showed better results in detecting significant long-range chromatin interactions from ChIA-PET experiments data than the preexisting tools.