==== Warning ====

**Do not share model weights, they are protected by a license!**

==== Starting a Job ====

**1.** Create a working directory with two subfolders: //input// and //output//

**2.** Place your JSON input file into the input folder ([[https://github.com/google-deepmind/alphafold3/blob/main/docs/input.md|input format documentation]]). Example:
<code>
{
  "name": "2PV7",
  "sequences": [
    {
     "protein": {
       "id": ["A", "B"],
       "sequence": "GMRESYANENQFGFKTINSDIHKIVIVGGYGKLGGLFARYLRASGYPISILDREDWAVAESILANADVVIVSVPINLTLETIERLKPYLTENMLLADLTSVKREPLAKMLEVHTGAVLGLHPMFGADIASMAKQVVVRCDGRFPERYEWLLEQIQIWGAKIYQTNATEHDHNMTYIQALRHFSTFANGLHLSKQPINLANLLALSSPIYRLELAMIGRLFAQDAELYADIIMDKSENLAVIETLKQTYDEALTFFENNDRQGFIDAFHKVRDWFGDYSEQFLKESRQLLQQANDLKQG"
      }
    }
  ],
  "modelSeeds": [1],
  "dialect": "alphafold3", 
  "version": 1
}
</code>

**3.** Create a Slurm job from [[https://portal.darwin.bioeng.ru/pun/sys/myjobs/workflows/new|Templates]]

**4.** Adjust the job script (for more details see [[portal_guides:slurm|Slurm Guide]])

**5.** Start the Job

==== Prediction Size Estimate (approximate) ====

| **Molecule type**  | **What 1 token represents**                   | **24 %%GB%% VRAM estimate** | **48 %%GB%% VRAM estimate** |
| Protein        | 1 amino acid residue                      | ~2,800 residues     | ~4,000 residues     |
| DNA / RNA      | 1 nucleotide                              | ~2,800 nt           | ~4,000 nt           |
| Ligand         | 1 atom                                    | ~2,800 atoms        | ~4,000 atoms        |
| Ion / cofactor | 1 atom / entity                           | ~2,800 entities     | ~4,000 entities     |
| Mixed complex  | Sum of all tokens across chains + ligands | ~2,800 tokens total | ~4,000 tokens total |