search

xiongxslab / Translatomer

MIT License
13 stars 2 forks source link

Available trained model weights (ckpt files) and processed data (bigwig files)? #2

Open kamisama0101 opened 3 weeks ago

kamisama0101 commented 3 weeks ago

Thank you for your amazing work! I want to try this model as soon as possible, but I find that some files in the tutorial are missing. Have the trained model weights (ckpt files) been released? If so, users as me can use this model without training. checkpoint = 'results/bigmodel_h512_l12_lr1e-5_epoch=38-step=746889.ckpt' model = TrainModule.load_from_checkpoint(checkpoint).to(device) model = model.eval() If the processed data (bigwig files) of 40 paired datasets is available, it's even better. It also means that other researchers can train or even refine the model without time-consuming pre-processing of raw FASTQ data. rna_bw_file = 'data/hg38/K562/GSE153597/input_features/rnaseq.bw' ribo_bw_file = 'data/hg38/K562/GSE153597/input_features/riboseq.bw'

kamisama0101 commented 3 weeks ago

Sorry for my carelessness... Now I have located the Example ckpt and bw files in Example data and pretrained Translatomer model (zenodo.org). Besides, I would like to ask if it is appropriate to publicly release the processed data (bigwig files) of all 40 paired datasets?

shtoneyan commented 2 weeks ago

Just to add to this, would it be possible to also request the sequence only model? Thank you!