search

bioinfomaticsCSU / deepsignal

Detecting methylation using signal-level features from Nanopore sequencing reads
GNU General Public License v3.0
108 stars 21 forks source link

Example training and validation data #53

Closed akk01 closed 3 years ago

akk01 commented 3 years ago

Can you also put example training and validation data, which will be really helpful in understanding input file.

PengNi commented 3 years ago

Hi @akk01 , to train a model, #13 , #10 , #7 , #32 and #50 would be heplful. The format of training and validation file is as the uploaded file. human.albacore.pass.tem.CpG.signal_features.bn17_sn360.train.sample.txt

Thanks for your interest of deepsignal.

Best, Peng

akk01 commented 3 years ago

Great! Thanks a lot.

akk01 commented 3 years ago

Hi Peng, currently DeepSignal accepts a specific input format for training and validation, which are generated by deepsignal extract. I am interested in using DeepSignal for my dataset, which is NxN numeric table and has two targets - modified sites or 1 and unmodified sites or 0. Please advise me if that can be done without making a big changes in the code.

PengNi commented 3 years ago

@akk01 , sorry, I don't quite understand what a NxN table means.

If you have features in different format, the architeture of the DNN model may need to be changed too,

Best, Peng

akk01 commented 3 years ago

Screen Shot 2020-09-03 at 2 14 16 PM This a dummy data format for which I want to use DeepSignal.

PengNi commented 3 years ago

Hi @akk01 , DeepSignal doesn't accept your data format. It seems that there are only sequence features in your data. I suggest that you design a new DNN model for your data.

Best, Peng