In this ReMap 2022 human release we have manually curated and annotated 10,619 ChIP-seq experiments, retained after quality control 8,103 datasets, and in 2020 one large ChIP-exo experiment (GSE78099) were included from Imbeault M. et al Nature 2017. We applied our pipeline for both type of data, however most postprocessing steps applied by Imbeault M. et al Nature 2017 where applied to the ChIP-exo data. We processed another recent large ChIP-exo experiment (GSE151287, n=1259 datasets, from Lai W.K.M et al bioRxiv 2021) which was however not retained as the peak numbers identified after QCs were inconsistent with the publication.

After consistent peak calling and stringent QCs, we identified a total of 182 million peaks bound by transcriptionnal regulators (including 990,476 peaks from ChIP-exo). These numbers include overlapping sites for identical TRs which were studied in various conditions. To address this we merged overlapping TR peaks for similar TR obtaining a catalog of XX million non-redundant peaks.

As not every ChIP-seq datasets are equal in terms of quality, we used four different metrics based on ENCODE ChIP-seq guidelines to retain high quality datasets for downstream analyses. First we used the normalized strand cross-correlation coefficient (NSC) which is a normalized ratio between the fragment-length cross-correlation peak and the background cross-correlation, and the relative strand cross-correlation coefficient (RSC), a ratio between the fragment-length peak and the read-length peak to exclude low quality datasets. We also used the fraction of reads in peaks (FRiP) and the number of peaks identified in each dataset to filter datasets.

Dataset(s) are plotted in a 2D vizualization with NSC and RSC as x- and y-axis, colours highlight the datasets conserved (green) or excluded (red) from the catalogue of binding sites.

In 2022 two sets of filters were added in our QC steps. The first filter consist of removing peaks those length are outside set cutoffs. In Human we defined the range betwen 50bp and 1.5kb. For each species the upper cutoff was selected based on the percentage of peaks, we searched the length under for which we have 98% of catalogue, so that we removed peaks above this cutoff. The second filter consist of removing datasets those peaks numbers are greater than two times the number of annotated genes (See Ensembl Human gene annotation statistics). As of late 2021, about 40,000 coding an non-coding genes are identified, giving us a cutoff of maximum 80,000 peaks per datasets.