Peaks of transcription factor occupancy ("optimal peak set") from ENCODE ChIP-seq datasets were clustered using the UCSC hgBedsToBedExps tool. Scores were assigned to peaks by multiplying the input signal values by a normalization factor calculated as the ratio of the maximum score value (1000) to the signal value at one standard deviation from the mean, with values exceeding 1000 capped at 1000. This has the effect of distributing scores up to mean plus one 1 standard deviation across the score range, but assigning all above to the maximum score. The cluster score is the highest score for any peak contributing to the cluster.
The raw data for the ENCODE3 TF Clusters track can be accessed from the - + Table Browser or combined with other datasets through the Data Integrator. This data is stored internally as a BED5+3 MySQL table with additional metadata tables. For automated analysis and download, the encRegTfbsClusteredWithCells.hg38.bed.gz track data file can be downloaded from our downloads server, which has 5 fields of BED data followed by a comma-separated list of cell types. The data can also be queried using the JSON API or the Public SQL server.
Thanks to the ENCODE Consortium, the ENCODE ChIP-seq production laboratories, and the ENCODE Data Coordination Center for generating and processing the TF ChIP-seq datasets used here. The ENCODE accession numbers of the constituent datasets are available from the peak details page.