Mouse embryonic stem cell data processing

To evaluate the utility of Chromoformer for species other than human, we processed the ENCODE reference epigenome of ES-Bruce4 mouse embryonic stem cell (mESC) line from its raw histone ChIP-seq reads (Supplementary Table ²). To be consistent with human data, the processing pipeline followed that of Roadmap Epigenomics Project as described below. After downloading FASTQ files, histone ChIP-seq reads were first aligned to mm9 reference genome using bwa v0.7.17-r1188⁵⁰. To normalize the effect of read length, each aligned read was then truncated up to 36 bp. Also, the read depths were normalized by subsampling the read alignment up to 3 million reads. Processed alignments were converted to genomewide read depth signals using bedtools genomecov. Besides, to determine the promoter-pCRE interactions that are used for Chromoformer training, we used the normalized interaction frequencies from publicly available Hi-C interaction matrices of mESC⁴¹. The bulk RNA-seq gene expression profile for ES-Bruce4 cells was also obtained from ENCODE under file accession ENCFF166EXS [https://www.encodeproject.org/experiments/ENCSR000CGU]. Gencode vM1 gene annotation was used to determine the transcription site and promoter region for each gene.