Fig. 3

Runtime comparison of alignment methods. a The time required to process 3005 cells from mouse brain cell dataset [7] in core hours is shown here. The time taken for read alignment with Bowtie and HISAT is much larger than the time taken for kallisto pseudoalignment (which is used by our method to obtain the transcript-compatibility counts). kallisto pseudoalignment and HISAT were run on 32 cores. Bowtie and Word Count were each timed on 1 core with 10 randomly selected cells. The bars shown here are estimates obtained by multiplying these times by 300.5. Because we do not account for the overhead associated with parallelizing, the Bowtie and Word Count estimates are lower bounds on their runtimes in practice. After preprocessing the UMIs, each of the 5,914,602,849 single-end reads in the dataset were less than 50 bp long. b The time required to process 271 cells of the dataset of [12] in core hours is shown here. As before, the time taken for read alignment with HISAT is significantly larger than the time taken for kallisto pseudoalignment. Both methods were run on 32 cores. The dataset has 814,344,693 paired-end reads, and each mate in a pair is 100 bp long