search

kundajelab / deepbind

I have put my modified version of the deepbind code here.
BSD 3-Clause "New" or "Revised" License
8 stars 10 forks source link

Score threshold selection #2

Open elenichri opened 7 years ago

elenichri commented 7 years ago

Hello, it's me again. I have a question on how you select a threshold for the deepbind score. I see in your example with 4 models and 4 sequences, that the good models have a high, positive value and the rest have a negative value. I ran deepbind on my own sequences and got many different scores, negative and positive. Many are in the range [0,1]. I want to select a score to define the 'good' models for each sequence. I read the supplementary file and it seems that you calculate for each motif (k-mer) a Z-score and E-score and an AUC, based on the k-mer's intensity. Is this how you determine the significant scores?

Thank you very much! Eleni

imk1 commented 7 years ago

Hi Eleni,

You should ask the authors of the original deepbind paper. I just copied their code and made a few minor tweaks.

Irene

elenichri commented 7 years ago

Ok Irene, I will. Thank you very much. Best regards, Eleni