MunGLER utilizes Large Language Models (LLMs) to unravel sample metadata from DNA sequencing data within the Sequence Read Archive.

• Visual Concept
• Installation
• Hardware requirements
• Usage
• Rationale
• Contributing
• License

First clone this repository

git clone https://github.com/fbcorrea/mungler.git

Create a conda environment. I recommend using Mamba.

mamba create -n mungler

Install python packages using pip

pip install -r requirements.txt

To effectively run this pipeline, it is strongly recommended to utilize a High-Performance Computing (HPC) system, ideally equipped with at least one recent high-end GPU card, such as the NVIDIA A100. This ensures optimal performance and reliability of the results.

Hardware requirements will be associated with the model used. Roughly, the larger the model, more memory it will require. If you chose to run Llama2 models from Meta, smaller models require 8GB vRAM while bigger up to 64GB vRAM. Other large models like Mixtral87b will require around 100GB vRAM.

Alternatively, some services allow users to pay per hour of process execution such as Amazon SageMaker (Amazon Web Services), Google Cloud AI Platform (Google Cloud Platform) and Azure Machine Learning (Microsoft Azure).

If you followed the installation instructions, you must activate your Conda environment to run the mungler main script.

conda activate mungler

In the first run, we recommend to run a test run with a toy dataset and tiny models.

mungler --acc-id SAMN06512631

If everything went well, you should get the following message:

Starting mungler test
Creating output
PASSED

Running mungler on a list of accession numbers. Any SRA, ENA, Biosample, Bioproject are valid. Invalid ids will be saved on a text file.

mungler 
    --acc-list list-of-ids.txt
    --output results.txt
    --invdir invalid-ids.txt

mungler takes as input accession numbers from SRA, access its metadata and tries to fill metadata fields

Before:

After:

Guidelines on how to contribute to the project.

Information about the project's license.