MTS-VAE

THE PAPER!

This repository accompanies the work "Design of diverse, functional mitochondrial targeting sequences across eukaryotic organisms using Variational Autoencoders". This work utilizes local software packages of TargetP 2.0 and DeepLoc 2.0 for synthetic data generation and analysis of artificial peptides, respectively. It also employs s4pred and modlamp for analyzing the characteristics of generated peptides, while UniRep embeddings are used for sampling.

Datasets curated for training the VAE models and analysis are made available on Zenodo: https://zenodo.org/records/14590156. Upload the data folders in MTS and Dual directories to run the provided scripts.

Model:

Install the requirements using:

conda env create -f environment.yml
conda activate mtsvae

Note: This setup was tested in a Linux environment. For smooth installation and best results, we recommend using Miniconda on Linux (https://docs.anaconda.com/miniconda/).

For model training, run:

python MTS/scripts/train.py

To generate artificial mitochondrial targeting sequences, run:

python MTS/scripts/generate.py

To sample artificial MTS near the cluster center (Note: For new VAE-generated MTSs, you need to update the fasta file for artificial sequences on line 97 and obtain UniRep embeddings (.npz) files):

python MTS/scripts/sample.py

To train the Dual-VAE model, run:

python Dual/scripts/train.py

To generate dual targeting sequences using latent interpolation, run:

python Dual/scripts/interpolate.py

Replicate the analysis/data processing performed in the paper using:

python MTS/analysis/distribution.py
python MTS/analysis/seq_identity.py
python MTS/analysis/deeploc2.py
python MTS/analysis/characteristics.py
python MTS/analysis/organism.py


python Dual/analysis/preprocess.py
python Dual/analysis/split.py
python Dual/analysis/deeploc2.py
python Dual/analysis/characteristics.py

Reference

If you using these scripts, please cite us:

Boob, Aashutosh Girish, et al. "Design of diverse, functional mitochondrial targeting sequences across eukaryotic organisms using variational autoencoder." bioRxiv (2024): 2024-08.