1.4 Input File Types in ArchR
ArchR can utilize multiple input formats of scATAC-seq data which is most frequently in the format of fragment files and BAM files. Fragment files are tabix-sorted text files containing each scATAC-seq fragment and the corresponding cell ID, one fragment per line. BAM files are binarized tabix-sorted files that contain each scATAC-seq fragment, raw sequence, cellular barcode id and other information. The input format used will depend on the pre-processing pipeline used. For example, the 10x Genomics Cell Ranger software returns fragment files while sci-ATAC-seq applications would use BAM files. ArchR uses “scanTabix” to read fragment files and “scanBam” to read BAM files. During this input process, inputs are chunked and each input chunk is converted into a compressed table-based representation of fragments containing each fragment chromosome, offset-adjusted chromosome start position, offset-adjusted chromosome end position, and cellular barcode ID. These chunk-wise fragments are then stored in a temporary HDF5-formatted file to preserve memory usage while maintaining rapid access to each chunk. Finally, all chunks associated with each chromosome are read, organized, and re-written to an Arrow file within a single HDF5 group called “fragments”. This pre-chunking procedure enables ArchR to process extremely large input files efficiently and with low memory usage, enabling full utilization of parallel processing.