Currently a lot of annotation data is loaded. This takes quite some time in the beginning of the task.
Since not a lot of the annotation is used (pysam is going to be used int he future), it's better to instead use gcsfuse and stream the bits we need. In the vast majority of cases, this should drastically reduce runtime.
TODO
First, change annotations to once again load themselves using tabix, this time pysam (but make sure it's efficient enough - probably better to load some ranges instead of single variants)
Second, preferably change annotations to inherit some sort of AnnotationDAO, and make all of them to behave well. Preferably nothing like the current shitshow with functions and loading stuff late etc.
Third, add gcsfuse to wdl, and everything should "just work"
Currently a lot of annotation data is loaded. This takes quite some time in the beginning of the task. Since not a lot of the annotation is used (pysam is going to be used int he future), it's better to instead use gcsfuse and stream the bits we need. In the vast majority of cases, this should drastically reduce runtime.
TODO