promoterai 1.0rc3

Predict the impact of promoter variants on gene expression

PromoterAI

This repository contains the source code for PromoterAI, a deep learning model for predicting the impact of promoter variants on gene expression, as described in Jaganathan, Ersaro, Novakovsky et al., Science (2025).

PromoterAI precomputed scores for all human promoter single nucleotide variants are freely available for academic and non-commercial research use. Please complete the license agreement; the download link will be shared via email shortly after submission. Scores range from –1 to 1, with negative values indicating under-expression and positive values indicating over-expression. Recommended thresholds are ±0.1, ±0.2, and ±0.5.

Installation

The simplest way to install PromoterAI for variant scoring is via:

pip install promoterai

For model training or to work directly with the source code, install PromoterAI by cloning the repository:

git clone https://github.com/Illumina/PromoterAI
cd PromoterAI
pip install .

PromoterAI supports both CPU and GPU execution, and has been tested on H100 (TensorFlow 2.15, CUDA 12.2, cuDNN 8.9.7) and A100 (TensorFlow 2.13, CUDA 11.4, cuDNN 8.6.0) GPUs.

Variant scoring

To score variants, organize them into a .tsv file with the following columns: chrom, pos, ref, alt, strand. If strand cannot be specified, create separate rows for each strand and aggregate predictions. Indels must be left-normalized.

chrom	pos	ref	alt	strand
chr16	84145214	G	T	1
chr16	84145333	G	C	1
chr2	55232249	T	G	-1
chr2	55232374	C	T	-1
chr1	64918	T	TGG	1
chr1	64918	TAA	T	1

Download the appropriate reference genome .fa file, and run the following command:

promoterai \
    --model_folder path/to/model_dir \
    --var_file path/to/variant_tsv \
    --fasta_file path/to/genome_fa \
    --input_length 20480

Scores will be added as a new column labeled score, with the output file named by appending the model folder’s basename to the variant file name.

Model training

To begin, download the appropriate reference genome .fa file and regulatory profile .bigWig files. Organize the .bigWig file paths and their corresponding transformations into a .tsv file, where each row represents a prediction target, with the following columns:

• fwd: path to the forward-strand .bigWig file
• rev: path to the reverse-strand .bigWig file
• xform: transformation applied to the prediction target

fwd	rev	xform
data/bigwig/ENCFF245ZZX.bigWig	data/bigwig/ENCFF245ZZX.bigWig	lambda x: np.arcsinh(np.nan_to_num(x))
data/bigwig/ENCFF279QDX.bigWig	data/bigwig/ENCFF279QDX.bigWig	lambda x: np.arcsinh(np.nan_to_num(x))
data/bigwig/ENCFF480GFU.bigWig	data/bigwig/ENCFF480GFU.bigWig	lambda x: np.arcsinh(np.nan_to_num(x))
data/bigwig/ENCFF815ONV.bigWig	data/bigwig/ENCFF815ONV.bigWig	lambda x: np.arcsinh(np.nan_to_num(x))

In addition, create a .tsv file listing the genomic positions of interest, with the following columns: chrom, pos, strand.

chrom	pos	strand
chr1	11868	1
chr1	12009	1
chr1	29569	-1
chr1	17435	-1

After preparing these files, run preprocess.sh with the paths to the genome .fa file, the profile and position .tsv files, and an output folder for writing the generated TFRecord files. For multi-species training, run the preprocessing step separately for each species. Next, run train.sh, specifying the TFRecord folder(s) and an output folder for saving the trained model. After training, run finetune.sh using the trained model as input. The fine-tuned model will be saved in a new folder with _finetune appended to the original model folder name.

Contact

• Kishore Jaganathan: [e-mail]
• Gherman Novakovsky: [e-mail]
• Kyle Farh: [e-mail]