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

The original FAST5 file looked for 6mA loci #55

Closed chenchen-eng closed 3 years ago

chenchen-eng commented 3 years ago

Hello, teacher, I hope to find 6mA from the original FAST5 file. How should I set up my training set?

PengNi commented 3 years ago

Hi @chenchen-eng ,

Currently deepsignal doesn't not perform well in 6mA detection. So for 6mA detection, I suggest that you try tombo or megalodon from nanoporetech.

Best, Peng

chenchen-eng commented 3 years ago

What I have is a fungus sample. If I look for 5mC, can I directly use your model? What should I do if I retrain?Thank you teacher

PengNi commented 3 years ago

@chenchen-eng , because the model provided deepsignal are trained by using human reads, retrain a model using reads of fungus sample is better way.

To train a model, high-confidence sites should be selected (from bisulfite sequenceing maybe). You can check other related issues ( #13 , #10 , #7 , #32 and #50).

If you don't have high-confidence sites, using the provided model of deepsignal is suggested. But the best performance of 5mC detection in your sample is not guaranteed.

Best, Peng

chenchen-eng commented 3 years ago

Thank you, teacher. Which model should 5mC use?

PengNi commented 3 years ago

@chenchen-eng , I suggest that you use the latest version of deepsignal and the model model.CpG.R9.4_1D.human_hx1.bn17.sn360.v0.1.7+.tar.gz.

You can download the model from google drive, and then uncompress it. Note that the model is for 5mCpG detection only.

Best, Peng