Hi David, I think it would be interesting to use an ensemble approach to decrease false positives in the samples. I think [MetaMeta](https://github.com/pirovc/metameta) could be useful.