yeast->human

[animesh]

what would be the recommended way to move the "dna repair" demo from yeast to human?

[michaelkyu]

This would be a non-trivial task because the current DCell model was trained on yeast. You would need to train a new DCell model using data in human cells.

[animesh]

ok, but what would it need in terms of input files? i have some free gpu time, can try to train, if could you tell me where to start from :)

[majianzhu]

1. Prepare an ontology file to restrict the neural network structure. The file needs to have the same format as the example.
2. Prepare a gene list to represent the gene deletions. Currently for memory issue, we only allow binary output. If you have any special real value input, please send me the file and I can help to adjust the code.
3. Tune the hyper-parameters based on your data.
4. Need some outside data to validate the state learned for each term as shown in the paper.