search

yasin-uzun / MAPLE.1.0

An R package for predicting gene activity level for single cell DNA Methylation data
1 stars 2 forks source link

Code option to re-train the Maple framework on new data #2

Open JiayiLi21 opened 1 month ago

JiayiLi21 commented 1 month ago

Hi Yasin,

I am trying the Maple package on new data, and just wondering is there any code support to re-train and test the model on new dataset in general.

Thank you!

yasin-uzun commented 1 month ago

Hi Jiayi,

For training, you need to have a true single-cell multiomics dataset. More speficially, it is necessary to have gene expression and DNA methylation from the same cells via a joint assay protocol such as scM&T-seq or scNMT-seq. Do you have such a dataset for training?

JiayiLi21 commented 1 month ago

Hi Yasin,

Our data does have paired measurement of DNA methylation and RNA expression from the same cell

yasin-uzun commented 1 month ago

Sounds good. The reason that I did not include the training component in the package was that for most single-cell multi-omics datasets available by the time, cell labeling was not clean and clear. Not all the cells were true multi-omics (some of them had RNA but not DNA and vice versa). Same cells were named with different conventions in two modalities and so on. I thought it would be a quite tedious process for any user to preprocess and clean these datasets and train the models. Hence, I had shared the pretrained models instead.

I will try to retrieve the training scripts and share them among the files.