Very low frequency genes and mutations are more likely to be associated with pheno S than with pheno R. To deal with this issue we've implemented empirically trained frequency thresholds below which, while still reporting the gene or variant we do not predict a resistant phenotype.
In the PR there are also changes to the models to use coverage rather than arrival rate in the Pois models. Though a fundamental change it primarily effects the confidence of calls rather than which model is best.
mecC was also added to the staph panel and L449X removed from TB.
Additional tests have been added for gene and mutation models.
This branch has been tested on the full TB and Staph Training and Validation sets.
Very low frequency genes and mutations are more likely to be associated with pheno S than with pheno R. To deal with this issue we've implemented empirically trained frequency thresholds below which, while still reporting the gene or variant we do not predict a resistant phenotype.
In the PR there are also changes to the models to use coverage rather than arrival rate in the Pois models. Though a fundamental change it primarily effects the confidence of calls rather than which model is best.
mecC was also added to the staph panel and L449X removed from TB.
Additional tests have been added for gene and mutation models.
This branch has been tested on the full TB and Staph Training and Validation sets.