Evan Cofer

This may be good to mention. I don't think abstraction necessarily means ignoring the theory and caveats of a set of algorithms. Given the effort that goes into designing and...

It is probably worth mentioning dropout ([discussed here](https://www.cs.toronto.edu/~hinton/absps/JMLRdropout.pdf)) and weight decay, both of which are extremely good for limiting overfitting.

I think something can be learned about the problem by determining which models perform well, right? For instance, the fact that [FactorNet](https://www.biorxiv.org/content/early/2017/06/28/151274) benefits from information such as DNase-seq data says...

There is another recent review ( https://doi.org/10.1038/nbt.4233 ) that I found to be relevant. Perhaps we could list these works in one place, or would it be better to just...

@agitter That is a nice touch. I think it could be interesting to implement a similar notebook for this paper, and show a sort of step-by-step DL application that uses...

It is also good to think about the potential applications that users might find for your model, and then validate/invalidate those applications if possible. For instance, validating variant effect prediction...

A nice corollary to this might be to favor convolutional networks over recurrent networks when determining model feasibility for sequential data (e.g. DNA sequence). Many features of CNNs, such as...

@jmschrei I would disagree, since the use of these models is not limited to identification of regulatory patterns. For instance, Basenji/ExPecto modeled gene expression from regulatory signals, a task for...

@jmschrei I agree with that sentiment about rigorous statistical evaluation. I would also go a step further and say that claims of novel biological mechanisms and phenomena should require experimental...

To many, one of the attractive features of DL is that it offers a high degree of flexibility (with respect to input feature representation and the tasks that it can...