Updated 6/21/2022 IntroductionAssuming you're familiar with the previous install, we'll go right into preparing a new Conda environment. Requires: OSCER Account Miniconda Differences with the previous install: - Alphafold v.2.2.0 (Multimer model weights updated; input flag changes) - Python 3.9 - Cudatoolkit 11.2.2 - Cudnn 8.1 - Tensorflow 2.5.0 - Updated database directories on OSCER Installation1. Log into OSCER, and in your home directory, download the environment file (right-click link, copy, and paste into command line). The YML has all required dependencies for Alphafold. wget https://cc-ats.weebly.com/uploads/7/5/3/4/75345355/environment.yml 2. Create the new Conda environment. conda env create -f environment.yml 3. In your Programs/ folder, download Alphafold. git clone https://github.com/deepmind/alphafold.git 4. Add the stereochemical properties file to Alphafold. cd alphafold/alphafold/common wget http://cc-ats.weebly.com/uploads/7/5/3/4/75345355/stereo_chemical_props.txt 5. Patch OpenMM, and you're good to go! cd ~/Programs/miniconda3/envs/af2.2.0/lib/python3.9/site-packages/ patch -p0 < ~/Programs/alphafold/docker/openmm.patch cd Testing Tensorflow, GPUs, etc.6. Download this script to your home directory, and submit it. The purpose of this script is to see if our installation was correct, namely, can TensorFlow recognize the GPU. Be sure to change the user directory for your Conda (/home/van/ is mine). wget https://cc-ats.weebly.com/uploads/7/5/3/4/75345355/findcuda.slurm sbatch findcuda.slurm You should see the following outputs (<job-number>.out) with either nodes c301 or c302. At the time of this writing node c307 is not usable. 2.5.0 Num GPUs: 2 [PhysicalDevice(name='/physical_device:GPU:0', device_type='GPU'), PhysicalDevice(name='/physical_device:GPU:1', device_type='GPU')] Additionally, both the working nodes have 2 GPU cards each. So with some additional scripting, it is possible to run 2 predictions with one job. If you would like help getting this set up, please send us an email. Running Alphafold Predictions8. You will need two files: a) the input amino acid sequence (fasta) b) SLURM script (monomer script, complex script) The rest of this post will follow the structure prediction of Bax. 9. Make a new directory, and input amino acid sequence. mkdir bax cd bax/ touch bax.fasta echo " >bax MDGSGEQPRGGGPTSSEQIMKTGALLLQGFIQDRAGRMGGEAPELALDPVPQDASTKKLSECLKRIGDELDSNMELQRMIAAVDTDSPREVFFRVAADMFSDGNFNWGRVVALFYFASKLVLKALCTKVPELIRTIMGWTLDFLRERLLGWIQDQGGWDGLLSYFGTPTWQTVTIFVAGVLTASLTIWKKMG" >> bax.fasta 10. Modify the SLURM script with your Conda and Alphafold program. 11. Change "INPUT" to "bax." sed -I "s/INPUT/bax/" monomer.slurm 12. Submit the script, and your prediction should be back within 24 hours! Notes on Alphafold Multimer Predictions (Complexes)Multimer predictions uses a few different databases. This is the biggest difference between the monomer and complex SLURM scripts.
Additionally, to run complex prediction, you just need to include additional proteins to the input amino acid sequence, separated by ">". Examples of such can be found on the Alphafold GitHub.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |