Compared the seed set calculation algorithm differences between the perl and python versions.

Method Comparison

Found some pretty substantial differences. I use the genome of the intracellular pathogen Buchnera aphidicola APS to benchmark the code as a comparison against Borenstein’s original paper. When I apply the perl code that was originally published on, I get the correct number of seeds at 68. However, when I use the updated python code, I calculate only 16.
These calculations are both done with identical parameters of a 5 node minimum per component and a minimum confidence of 0.2 per seed. Emailed Rogan Carr from their lab to hopefully shed some light on what’s going on. More to come soon…

Repeating in vitro experiments

Deriving growth curves for each strain and redoing all competition experiments. Nutrient re-supplementation experiments are up next.