promoter_prediction

Promoter prediction

We compared performance of GENA-LM models to

1. BigBird https://papers.nips.cc/paper/2020/hash/c8512d142a2d849725f31a9a7a361ab9-Abstract.html
2. DNABERT https://pubmed.ncbi.nlm.nih.gov/33538820/

Dataset Preparation

Step 1. Download data

Original data was from EPDNew: https://epd.epfl.ch/EPDnew_select.php; note that the EPDnew database was recently moved here. We used EPDNew select tool to fetch human promoter sequences (hg38). Four different sequence lengths are used:

1. Length 300. From -249 to 50. Results in a file hg38_len_300.fa.txt
2. Length 2000. From -1000 to 999. Results in a file hg38_len_2000.fa.txt
3. Length 8000 (like in BigBird paper). From -5000 to 2999. Results in a file hg38_len_8000.fa.txt
4. Length 16000. From -8000 to 7999. Results in a file hg38_len_16000.fa.txt

Step 2. Create a dataset

Run the script `dataset_generator.py`` with fasta files obtained in previous step.

>> python dataset_generator.py
hg38_len_300.fa.txt

The script treats promoter sequences as positive targets and generates negative samples, following the same procedure as in DeePromoter paper. Results in:

hg38_promoters_len_300_dataset.csv

Step 3. Split to 5 folds

Run the dataset_fold_split.py script with csv files obtained from dataset generator

>> python dataset_fold_split.py
hg38_promoters_len_300_dataset.csv

Results in five csv files named from fold_1.csv to fold_5.csv and corresponding train/valid/test splits.