GenSyntax 🧬

Model

We release our fine-tuned genome inference model on Hugging Face:

👉 GenSyntax on HuggingFace

You can load it in Python as follows:

from transformers import AutoTokenizer, AutoModelForCausalLM

model_name = "shuaimin4588/GenoVerse"

tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name, device_map="auto")

Hardware

A single 4090 GPU is sufficient for model deployment and inference.

Data

We release our data on HuggingFace:

👉 Datasets on HuggingFace (This includes the complete test sets for each task, as well as the training data and test data for cell phenotype prediction.)

Installation

Clone this repository and install the required dependencies:

git clone https://github.com/your-repo/GenSyntax.git

cd GenoVerse

pip install -r requirements.txt

Inference Tasks

Plasmid Host Prediction

python Plasmid_host_identification.py \
    --model checkpoint \
    --input-json-paths test_data/gene_task1_test_1000_format.json

Gene Function Prediction

python Gene_function_prediction.py \
    --model checkpoint \
    --input-json-paths test_data/gene_task2_test_500_opts.json

Genome Assembly

python Genome_assembly.py \
    --model checkpoint \
    --input-json-paths test_data/gene_task3_test_500_contig3_format.json

Gene Essentiality Prediction

python Gene_essentiality_prediction.py \
    --model checkpoint \
    --input-json-paths test_data/gene_task4_test_1000_format.json

Derivation of minimal genomes

python minimal_genome_inference.py \
    --model checkpoint \
    --input-json-paths test_data/bacteria_chromosomes_9-mini.json