A modern genomics pipeline is a directed graph of computational stages, each with its own throughput ceiling — and the whole pipeline is only as fast as its slowest connected stage.
throughput_i = min(input_i, capacity_i * QC_pass_rate)
- Pipeline stages — steps a FASTQ file passes through: trimming, alignment (BWA), variant calling (GATK), annotation (VEP).
- Stage connectivity — how tightly downstream stages depend on upstream output — a linear vs branching DAG.
- Read throughput — sequencer output rate feeding the front of the pipeline.
- QC threshold — minimum quality score a read must clear before continuing downstream, discarding the rest.
Cloud genomics platforms (DRAGEN, nf-core) are engineered almost entirely around this bottleneck problem: alignment and variant calling are the computationally heaviest stages and get parallelized across nodes.