77bde5b09a106b8aa6746508bbf783bd97e251a8
examples/alphafold3.md
... | ... | @@ -1,12 +1,14 @@ |
1 | 1 | ## AlphaFold3 Example |
2 | 2 | |
3 | 3 | In order to run AlphaFold3 with SBGrid you must: |
4 | + |
|
4 | 5 | 1. Obtain the AF3 models |
6 | + |
|
5 | 7 | 2. Download the AF3 databases |
6 | 8 | |
7 | -### Obtain the AF3 models. |
|
9 | +### Obtain the AF3 models |
|
8 | 10 | |
9 | -SBGrid cannot distribute these. You must apply through this form: |
|
11 | +SBGrid cannot distribute the AF3 models. Currently, you must apply to obtain them yourself through this form: |
|
10 | 12 | |
11 | 13 | https://forms.gle/svvpY4u2jsHEwWYS6 |
12 | 14 | |
... | ... | @@ -16,7 +18,7 @@ You will receive a link to download the models as a zstd compressed file. Uncomp |
16 | 18 | |
17 | 19 | ### Download databases |
18 | 20 | |
19 | -The databases for AlphaFold3 (AF3) must be downloaded. |
|
21 | +The databases for AlphaFold3 (AF3) must be downloaded. The uncompressed files require ~1TB of space. |
|
20 | 22 | |
21 | 23 | AlphaFold provides a script for downloading these in the github repo: |
22 | 24 | |
... | ... | @@ -30,7 +32,7 @@ Since then AF3 has changed to use the uncompressed files in the mmcif tar file. |
30 | 32 | The resulting layout should look like this: |
31 | 33 | |
32 | 34 | ``` |
33 | -/programs//local/alphafold-3.0.0/databases/ |
|
35 | +/programs/local/alphafold-3.0.0/databases/ |
|
34 | 36 | ├── bfd-first_non_consensus_sequences.fasta |
35 | 37 | ├── compressed |
36 | 38 | │ ├── bfd-first_non_consensus_sequences.fasta.zst |
... | ... | @@ -63,11 +65,11 @@ We use the input json file given in the AF3 github readme; alphafold_input.json. |
63 | 65 | ``` |
64 | 66 | #!/usr/bin/env bash |
65 | 67 | |
66 | -#SBATCH --partition=bch-gpu |
|
68 | +#SBATCH --partition=my-gpu-partition |
|
67 | 69 | #SBATCH --gres=gpu:large:1 |
68 | 70 | #SBATCH --mem-per-cpu=24G |
69 | 71 | #SBATCH --time=04:00:00 |
70 | -#SBATCH --job-name=JV-AF3 |
|
72 | +#SBATCH --job-name=AF3-Example |
|
71 | 73 | |
72 | 74 | nvidia-smi --query-gpu=name,memory.total --format=csv |
73 | 75 | |
... | ... | @@ -78,7 +80,7 @@ nvidia-smi --query-gpu=name,memory.total --format=csv |
78 | 80 | ## Nov 11, 2024 |
79 | 81 | |
80 | 82 | ## Start SBGrid environment |
81 | -source /programs/biogrids.shrc |
|
83 | +source /programs/sbgrid.shrc |
|
82 | 84 | export ALPHAFOLD_X=3.0.0 |
83 | 85 | |
84 | 86 | export XLA_FLAGS="--xla_gpu_enable_triton_gemm=false" |
... | ... | @@ -86,9 +88,9 @@ export XLA_FLAGS="--xla_gpu_enable_triton_gemm=false" |
86 | 88 | export XLA_PYTHON_CLIENT_PREALLOCATE=true |
87 | 89 | export XLA_CLIENT_MEM_FRACTION=0.95 |
88 | 90 | |
89 | -PROJECT_DIR=/temp_work/ch199734/121503_alphafold3_3.0.0 |
|
90 | -MODEL_DIR=/home/ch199734/models |
|
91 | -DATABASE_DIR=/programs/local/biogrids/alphafold-3.0.0/databases |
|
91 | +PROJECT_DIR=/tmp_space/user123/alphafold3_3.0.0 |
|
92 | +MODEL_DIR=/home/user123/models |
|
93 | +DATABASE_DIR=/programs/local/alphafold-3.0.0/databases |
|
92 | 94 | |
93 | 95 | cd ${PROJECT_DIR} |
94 | 96 |