Introduction

This repository contains the files needed to recreate our group's project on m6A modification detection

Getting Started

To run some training / inference with the model discussed in our report, follow these instructions :

1. Start a new AWS ubuntu instance (provisioning a new instance avoids conflicts with previously set paths etc), ensure that it is at least a large instance type.
2. From the home directory, clone this repo : git clone https://github.com/g4ryy/DSA4262-frontasticfour.git
3. Enter the demo folder within our repo : cd DSA4262-frontasticfour/demo/
4. Running the model training / inference requires some setup. We automate this using a shell script.
   • To grant permissions for the script to run call : chmod +x setup_script.sh
   • To install all dependencies call : source ./setup_script.sh This may take a few minutes

Running Inference

A sample dataset has been provided to run a small prediction / inference demo. From within DSA4262-frontasticfour/demo/ do the following :

1. Enter the m6Anet folder : cd m6Anet/

2. To run the pre-trained model on the sample dataset, call : python3 run_inference.py ../inference_sample.json

The resulting csv file with the m6A modification scores will be placed in : DSA4262-frontasticfour/demo/inference_sample_results.csv

Predictions can be made on any dataset (with the same format) by changing the given datafile path. Call python3 run_inference.py -h for more details on the required input arguments.

Running Training (Optional)

There is no sample dataset provided to run the training of the model as the training data & label files required are too large to store on github. However if these files have been stored outside the repo it is still easy to do the training :

1. Ensure you are still in the DSA4262-frontasticfour/demo/m6Anet folder
2. Call python3 run_learner.py <path to data.json file> <path to data.info file>

The model training results such as the training loss & Validation loss at each epoch will be placed in new folder in the current directory. Call python3 run_learner.py -h for more details on required input arguments