Skip to content

Latest commit

 

History

History

DeepSTARR

DeepSTARR

Predict drosophila enhancer activity based on DNA sequence

Dataset can be downloaded from Stark lab, repo for original DeepSTARR CNN solution here

To download data run:

# FASTA files with DNA sequences of genomic regions from train/val/test sets
wget 'https://data.starklab.org/almeida/DeepSTARR/Data/Sequences_Train.fa'
wget 'https://data.starklab.org/almeida/DeepSTARR/Data/Sequences_Val.fa'
wget 'https://data.starklab.org/almeida/DeepSTARR/Data/Sequences_Test.fa'

# Files with developmental and housekeeping activity of genomic regions from train/val/test sets
wget 'https://data.starklab.org/almeida/DeepSTARR/Data/Sequences_activity_Train.txt'
wget 'https://data.starklab.org/almeida/DeepSTARR/Data/Sequences_activity_Val.txt'
wget 'https://data.starklab.org/almeida/DeepSTARR/Data/Sequences_activity_Test.txt'

No preporcessing is needed, we use train/valid/test as in the original dataset:

train/valid/test = 402296/40570/41186 samples

Finetuning on DeepSTARR

Set paths to the data, edit hyperparameters in example script finetune_deepstarr.sh and run training:

CUDA_VISIBLE_DEVICES=0 NP=1 ./finetune_deepstarr.sh