Confirmed the parameterization for dadi, and polyDFE.
For DFE-alpha, the prior code seems to have been off by a factor 2. This fix will bring inference and simulations closer together.
For GRAPES, I'm a bit confused. The papers suggest that the population-scaled selection coefficient should be 4 Ne times the unscaled coefficient. But the agreement between inference and simulations suggests it is just Ne times the unscaled coefficient. It's not explicit in the software manual what convention is followed, and grepping the source code didn't make it clear to me either.
Confirmed the parameterization for dadi, and polyDFE.
For DFE-alpha, the prior code seems to have been off by a factor 2. This fix will bring inference and simulations closer together.
For GRAPES, I'm a bit confused. The papers suggest that the population-scaled selection coefficient should be 4 Ne times the unscaled coefficient. But the agreement between inference and simulations suggests it is just Ne times the unscaled coefficient. It's not explicit in the software manual what convention is followed, and grepping the source code didn't make it clear to me either.