mireklzicar / asmac

A machine learning model for pairwise comparison of ribosomal RNA sequences.

AsMac

This is the multiprocessing implementation of AsMac, a machine learning framework designed for pairwise comparison of ribosomal RNA sequences. The following instruction is for Linux/MacOS.

1. Installation

If a python environment for Python 3.7 or later version has not been set up yet, please follow these steps to create a virtual environment.
Install and create a virtual environment for python3

sudo pip3 install virtualenv
python3 -m venv venv3

Activate the python virtual environment. Then, install packages.

source ./venv3/bin/activate
pip install -r requirements.txt

Build the cython executable by:

python setup_softnw.py build_ext --inplace

2. Distance estimation

Use -h or --help flags to get help text for the program.

python AsMac.py -h

Choose a model with the -m or --model flags, and run the script:

python AsMac.py -i path/to/input.fasta -o path/to/output.csv -m 16S-full

for example:

python AsMac.py -i ./data/testing_seq.fa -o ./out.csv -m 16S-full

AsMac takes input sequences written in fasta format and outputs a pairwise distance matrix in a CSV table. For example:

	seq_1	seq_2	seq_3
seq_1	0	0.2	0.3
seq_2	0.2	0	0.1
seq_3	0.3	0.1	0

3. Encoding sequence representations

The embedding.py script provided the demo for using AsMac as a pre-trained encoder.

4. Train new model (optional)

Preparing training and testing data

Download C++ library SeqAn or use the attached version to compile the C++ code for NW alignment.

unzip seqan.zip
g++ -I . -std=c++1z -o CppAlign/align CppAlign/main.cpp CppAlign/read_fasta.cpp

Then generate alignment distance results for the input sequences.

./CppAlign/align ./data/training_seq.fa 0

The alignment process is very time-consuming on a personal computer. The demo code uses the finished result: training_dist_prepared.txt

Run the demo code for training

jupyter notebook demo.ipynb

Reference

1. Jian Chen, Le Yang, Lu Li, Steve Goodison, and Yijun Sun. "Alignment-free Comparison of Metagenomics Sequences via Approximate String Matching." Bioinformatics Advances (2022).
   
2. Koide, Satoshi, Keisuke Kawano, and Takuro Kutsuna. "Neural edit operations for biological sequences." Advances in Neural Information Processing Systems 31 (2018).