This repository is no longer maintained. The latest version of AliNA will be available at https://github.com/XELARI-CORP/AliNA

AliNA

(ALIgned Nucleic Acids) is an RNA secondary structures prediction algorithm based on deep learning. It is used to predict the DotBracket notation of secondary structure based on primary sequence.

Installation

• Download the repository and change directory:

  git clone https://github.com/Arty40m/AliNA.git
  cd AliNA
  

• You need python version 3.8 or greater. We recommend using conda to install a particular python version in case your system is below 3.8.

• Create virtual environment:

  • Using pip:

  python3 -m venv Alina_venv
  . Alina_venv/bin/activate
  

  • Using conda:

  conda create --name Alina_venv python=3.10
  conda activate Alina_venv
  

• Install AliNA as an editable package:

  python3 -m pip install -e .
  

• PyTorch is required for AliNA to work. It is not installed automatically, since you have to specify your CUDA version and operating system. See the official install page for more detail. Stable PyTorch version for AliNA is 1.11. If GPU usage is not important for you, just try:

  python3 -m pip install torch==1.11
  

• Test AliNA:

  python3 -m pip install -e .[dev]
  pytest tests --verbose
  

Usage

AliNA can be used from console (CLI) and as a python package (API).

CLI

Quick example:

- RNA sequence
    alina -i GGAGCAUUACCCCCCCAAAACCCUGGGGAUACAGGGCCCA
- Fasta file
    alina -m file -i path_to_file.fasta -o path_to_prediction_file.out

alina [-h] [-m {seq,file}] -i <Sequence or Fasta file> [-o <Output file>] [-th <Threshold value>] [--gpu] [--skip-errors] [--no-warn]

-m, --mode - prediction mode: "seq" - for single sequence typed in the command. "file" - for multiple predictions from .fasta file.

-i, --input - input sequence or path to fasta file.

-o, --out - path to the output file for "file" mode. Default - Prediction_(input file name).

-th, --threshold - specifies the sensitivity threshold when processing predictions [0, 1]. The lower the threshold the more bonds will be included into the predicted structure. Unless specified explicitly, the default value is 0.5.

--gpu - run model using GPU if CUDA is available. Default - run on CPU only.

--skip-errors - Skip invalid sequences instead of raising SequenceError. Default - False.

--no-warn - Do not raise warning on invalid sequnce skip. Default - False.

-h, --help - show help message.

API

You can import AliNA as a python package and use its API to predict secondary structure in different forms - as a dot bracket or matrix of raw model probabilities. Arguments specified in command line can be passed into class constructor:

import alina

Alina = alina.AliNA(skip_error_data = False,
                    warn = True,
                    gpu = False)
                
struct = Alina.fold('UAGCGUAGGGGAAACGCCCGGUUACAUU')
struct
>>> .((([[.(((......))).)))]]...

struct = Alina.fold('UAGCGUAGGGGAAACGCCCGGUUACAUU', threshold = 0.7)
struct
>>> ..((.....(......)...))......

structs = Alina.fold(['CUCUACUUACGGCGCUAAUAAAAAAGCGUUACGUUUU', 'AAAAAAAAAAAAAUUUUUUUUUUUUU'])
structs
>>> ['........((((((((........)))))..)))...', '((((..................))))']

To get indexes of complementary pairs:

struct = Alina.fold('UAGCGUAGGGCCCGGUUACAUU')
struct
>>> ....((((.......))))...

alina.utils.get_na_pairs(struct)
>>> [(7, 15), (6, 16), (5, 17), (4, 18)]

To obtain probabilities and processed quantized matrix:

import matplotlib.pyplot as plt

s, p = Alina.fold('GUUUCGGUGGUCAUAGCGUAGGGGAAACGCCCGGUUACAUUUCGAACCCGGAAGCUAAGCCUUACAGC', 
              with_probs=True)
s
>>> ................(((((((.......((..((........))...))........)))))).).

plt.imshow(p)

q = alina.utils.quantize_matrix(p, threshold = 0.5)
plt.imshow(q)

Cite this work

• Nasaev, S., Mukanov, A. R., Kuznetsov, I. I., Veselovsky, A. V. (2023). AliNA – a deep learning program for RNA secondary structure prediction. Molecular Informatics. https://doi.org/10.1002/minf.202300113