Reads were aligned to the 1000 genomes phase 2 reference genome (hs37d5) which includes build GRCh37 and decoy sequences: (ftp://ftp-trace.ncbi.nih.gov/1000genomes/ftp/technical/reference/phase2_reference_assembly_sequence/hs37d5.fa.gz) using the Illumina Dragen 3.7.3 aligner. The ichorCN...
In this way, the instrumentation can be written as a two-pass procedure: first the instrumented bytecode is emitted with placeholders for the jump targets, and then the placeholders are patched with the final offsets of the jump targets. The algorithm is further explained in a long comment in...
however, I was unable to figure out how exactly (it is not monotonic). An easy way to test this is to start with a convnet without batchnorm, and start stacking batchnoms at an arbitrary position