🧬 SiamProm

Cyanobacterial promoter identification with Siamese network-based contrastive learning

Novel non-promoter generation · Siamese contrastive learning

Usage

Google Colab

Predict promoters from pasted or uploaded 81 bp sequences on a free Colab GPU, no installation needed.

Local

Environment

conda env create -f env_SiamProm.yaml
conda activate SiamProm

Inference

python predict.py --checkpoint weights/siamprom_phantom.pth --fasta seqs.fasta --output results.csv

Arguments

Argument	Description	Default
--checkpoint	Model checkpoint path	—
--fasta	Input FASTA file (81bp sequences)	—
--output	Output CSV path	—
--device	cpu or GPU index	0
--batch-size	Batch size	256
--threshold	Classification threshold	0.5

The output CSV contains columns: name, sequence, prediction, confidence.

Train from scratch

python train.py sampling=phantom

Options

Optional values for the parameter sampling are phantom(default), random, cds, and partial.

For more detailed instructions on parameter management and configuration usage, please refer to the Hydra documentation.

Citation

@article{yang2024recognition,
  title   = {Recognition of cyanobacteria promoters via Siamese network-based contrastive learning under novel non-promoter generation},
  author  = {Yang, Guang and Li, Jianing and Hu, Jinlu and Shi, Jian-Yu},
  journal = {Briefings in Bioinformatics},
  volume  = {25},
  number  = {3},
  pages   = {bbae193},
  year    = {2024},
  doi     = {10.1093/bib/bbae193},
}