We agreed to focus on rare performance for v1 of the paper, but we may be able to improve common performance by tackling some low hanging fruit, e.g.:
bin-specific priors for the region-type HMM backbone (@asmirnov239 is working on a branch)
changing MQ-based filtering of reads or annotation-based filtering of bins (@samuelklee showed that we are losing data in many common regions to these hard filters even before we get to inference)
We agreed to focus on rare performance for v1 of the paper, but we may be able to improve common performance by tackling some low hanging fruit, e.g.: