In light of the recent scandal Scott Reuben scandal never has it been more important to properly document how scientific results were achieved that they may be reproducible. While this is hard in large clinical trials one field where this should not be a problem is computer science particularly software oriented computer science. If a paper is published it should also have some sort of link where the software used can be found and step by step instructions of how the results were achieved.
I am currently working on prediction phosphorylation locations on proteins and was hoping to use a recent work as a gauge and starting point, but after 4 days of work I still can’t replicate the results that were published. This is a HUGE oversight. Given that the models can be trained and tested in under 10 minutes it should have taken me under an hour to get everything working but instead I’m sorting through their code making wild predictions as to what all the methods they have created mean. (Hint all the names are virtually the same.) That having been said the published results were very good and I hope I’m getting close to getting their software working. Soldiering on!