Please find the guidelines for installing required dependencies below.
Our install requires conda or mamba.
For installation of these, please follow our guidelines in the repository root README.md.
If you have conda or mamba, we provide a ready to use install script via
./install.sh
This will install all dependencies. The script was tested on Ubuntu 22.04.5 LTS ((Jammy Jellyfish).
We provide guidelines to install our environment on linux64 and osx64.
Run all following commands from the project root.
For manual istall, you can run
conda env create -f env/environment.yml
Or install via our conda-lock file (requires conda-lock)
conda-lock install --name synEvo --file env/conda-lock.yml --platform linux-64
Continue with the guidelines for all environments below.
We provide a conda-lock file for osx. You can install the environment with
conda-lock install --name synEvo --file env/conda-lock.yml --platform osx-64
activate the environment with
conda activate synEvo
Continue with the guidelines for all environments below.
Activate the envrionment with
conda activate synEvo
After that, run
pip install -e .
to install pip dependencies.
After that, you can clone RnaBench via
git clone https://github.com/automl/RnaBench.git
For SPOT-RNA, first create the external_algorithms directory
mkdir -p external_algorithms
and move there
cd external_algorithms
You can now clone SPOT-RNA via
git clone https://github.com/jaswindersingh2/SPOT-RNA.git
For downloading SPOT-RNA models, run
cd SPOT-RNA && wget https://www.dropbox.com/s/dsrcf460nbjqpxa/SPOT-RNA-models.tar.gz
and extract the models via
tar -xzvf SPOT-RNA-models.tar.gz
Then go back to the project root.
We provide a CPU runtime patch for RNAformer so one can also run it here without any further dependencies. However, this patch mainly serves demonstration purposes and we strongly recommend running RNAformer on GPU as desccribed below for any performance critique use cases.
You can use the infer_d branch of the original github repository of the RNAformer to produce secondary structure predictions using GPUs.
To do so, please follow the install instructions as desribed in the RNAformer repository: https://github.com/automl/RNAformer. The models used in our manuscript can be downloaded following the instructions in the README
We use the inference code of AlphaFold 3 to run our experiments.
Note that the AlphaFold 3 webserver at https://alphafoldserver.com/ currently does not support custom MSAs.
To install AlphaFold 3, we follow the instructions at https://github.com/google-deepmind/alphafold3.
NOTE that using AlphaFold 3 requires to agree to the terms of usage and requires to request the open weights.
For completeness, we also provide instructions for the usage of DSSR. We use X3DNA-DSSR to obtain secondary structure information from 3D structures. While not necessary for our pipeline, this step is currently required during evaluation of the 3D predictions.
However, X3DNA-DSSR is currently available under license and requires application. You can find more infomation here: https://x3dna.org/.
** We provide all DSSR json files for download for reproducing our results. There is no need to install DSSR in general to generate our synthetic homologous sequences! We only use it for secondary structure evaluation!**
We use US-align in our evaluation pipeline. We provide links and some calls for the installs of US-align. Please find detailed instructions on US-align at the US-align Website.
To compile US-align from scratch, you can directly download the USalign.cpp an compile with
g++ -static -O3 -ffast-math -o USalign USalign.cpp
Alternatively, you can also directly download a zip of the binaries for your OS.