Top X things to try when troubleshooting training in biology

[cgreene]

I'm not entirely sure that the framing of this is quite right yet, but during a seminar this week @laserson suggested that we might consider including a section of the paper that is more "strategies to try" oriented than what we have. I'm imagining somewhere in the discussion that references specific strategies that have been used (for example, data augmentation, adding structure to the NN, multi-task learning).

I'd be happy to start to try to craft a bit of what this could look like (maybe 5 strategies initially?), particularly if @laserson agrees on scope.

@evancofer : what is the current timeline for text updates?

[evancofer]

This seems interesting.

I haven't seen a lot of activity in the past month, so I've been looking over deep learning methods for variant prioritization (e.g. this) and RNA-binding protein activity prediction, as there seem to have been several recent papers/preprints on both that weren't covered in the previous review. I think some updates will stir up activity.

The timeline is unclear. I can name certain subsections that need updates, but it would be great to get feedback/discussion of others that we aren't familiar with. I am unsure how to bring this discussion to the deep-review beyond stirring up more general activity via my own revisions.

[AlexanderTitus]

I think an additional section on deep learning in epigenetics to complement the section on gene expression would be a good addition. I would be happy to provide an initial draft of the section.

[laserson]

That all sounds great to me @cgreene! I would also vote that it would include some higher-level view of the common threads across all the papers you've seen (analogous to the set of slides you had to tear through at the end of your talk).

[evancofer]

@AlexanderTitus This should be submitted as a separate issue.