locimend

locimend is a tool that corrects DNA sequencing errors using Deep Learning.

The goal is to provide a correct DNA sequence, when a sequence containing errors is provided.

It provides both a command-line program and a REST API.

Technologies

• Tensorflow
• Biopython
• FastAPI

Installation

This project uses Nix to ensure reproducible builds.

1. Install Nix (compatible with MacOS, Linux and WSL):

curl -L https://nixos.org/nix/install | sh

2. Clone the repository:

git clone https://git.coolneng.duckdns.org/coolneng/locimend

3. Change the working directory to the project:

cd locimend

4. Enter the nix-shell:

nix-shell

5. Install the dependencies via poetry:

poetry install

After running these commands, you will find yourself in a shell that contains all the needed dependencies.

Usage

Training the model

The following command creates the trains the Deep Learning model and shows the accuracy and AUC:

poetry run python locimend/main.py train <data file> <label file>

• : FASTQ file containing the sequences with errors
• : FASTQ file containing the sequences without errors

Both files must contain the canonical and read simulated sequences in the same positions (same row).

A dataset is provided to train the model, in order to proceed execute the following command:

poetry run python locimend/main.py train data/curesim-HVR.fastq data/HVR.fastq

Inference

A trained model is provided, which can be used to infer the correct sequences. There are two ways to interact with it:

• Command-line execution
• REST API

Command-line

The following command will infer the correct sequence, and print it:

poetry run python locimend/main.py infer "<DNA sequence>"

REST API

It is also possible to serve the model via a REST API, to start the web server run the following command:

poetry run api

The API can be accessed at http://localhost:8000, with either a GET or POST request:

Request	Endpoint	Payload
GET	/	Sequence as a path parameter (in the URL)
POST	/	JSON

For a POST request the JSON must have the following structure:

{"sequence": "<DNA sequence>"}