Basenji Features

Hi! 

Congrats on the great paper :)

I have a question regarding how the Basenji features were obtained. Basenji produces expression predictions across the sequence. To obtain the features which are used for training the model, do you only take the bin which aligns with the TSS, or do you also take surrounding bins and then perform some form of aggregation?

Thank you!