Data Request with 'train_in_3_mers' key in h5 dataset

[shashivish]

Hi ,

I am trying to use below dataset shared earlier but it does not has train_in_3_mers key in it.

https://cuhko365-my.sharepoint.com/:u:/g/personal/221041009_link_cuhk_edu_cn/EccXeRSaTDJHjBDrdD2m06ABKev_9XPNl75Ew8bNlwLIEQ?e=ZibL4V

when I try to train my model with embeddings , it fails as above h5 files shared does not has train_in_3_mers in it.

Will you be able to share dataset with train_in_3_mers for custom training ?

Thank You
Shashi

[Tsedao]

Hi there,

Because of the large file size, we can't provide the sequence embeddings in the .h5 file.

If you want to train the embedding models, you can easily derive and play with your own 'train_in_3_mers' vectors using the word2vec model in the Gensim package.

Best

[shashivish]

Thank you for your response.

One quick question , I was looking at train dataset from h5 file , I was wondering how do we know where is modification in RNA sequence and what type of modification it is to train embeddings. Any pointers on this ?

[Tsedao]

For the dataset we provided, we have already separated the sequences into 12 different classes and the modification lies exactly in the middle point of the provided RNA sequence. In other words, during the training process, we are learning whether there exists a modification in the middle point of that sequence or not.

[shashivish]

Thank you very much. I will try to spend more time to understand it better.