AlphaFold 3

import google.auth
from google.auth.transport.requests import AuthorizedSession
import vertexai
from vertexai import model_garden

PROJECT_ID = "YOUR_PROJECT_ID"
LOCATION = "us-west1"
MODEL_ID = "google/alphafold3@v3_0_4"
MACHINE_TYPE = "a3-highgpu-1g"

vertexai.init(project=PROJECT_ID, location=LOCATION)

# 1. Deploy Model Garden OpenModel to Dedicated Endpoint
af3_model = model_garden.OpenModel(MODEL_ID)
endpoint = af3_model.deploy(
    endpoint_display_name="af3-dedicated-ep",
    model_display_name="af3-on-mg",
    machine_type=MACHINE_TYPE,
    accelerator_type="NVIDIA_H100_80GB",
    accelerator_count=1,
    reservation_affinity_type="ANY_RESERVATION",
    use_dedicated_endpoint=True,
    accept_eula=True,
    min_replica_count=1,
    max_replica_count=1,
    serving_container_deployment_timeout=3600,
)

# 2. Update inference timeout to 3,600 seconds
credentials, _ = google.auth.default(
    scopes=["https://www.googleapis.com/auth/cloud-platform"]
)
session = AuthorizedSession(credentials)
url = f"https://{LOCATION}-aiplatform.googleapis.com/v1/{endpoint.resource_name}:update"
payload = {
    "endpoint": {
        "name": endpoint.resource_name,
        "clientConnectionConfig": {
            "inferenceTimeout": {
                "seconds": 3600
            }
        }
    }
}
response = session.post(url, json=payload)
response.raise_for_status()
print(f"Endpoint Resource Name: {endpoint.resource_name}")

POST https://HOST/v1/projects/PROJECT_ID/locations/LOCATION/endpoints/ENDPOINT_ID:predict

{
  "instances": [
    {
      # The AlphaFold 3 input JSON - see the input documentation at
      # https://github.com/google-deepmind/alphafold3/blob/main/docs/input.md
    }
  ],
  "parameters": {
    "output_dir": "string",
    "dry_run": boolean,
    "run_data_pipeline": boolean,
    "force_output_dir": boolean,
    "resolve_msa_overlaps": boolean,
    "max_template_date": "string",
    "conformer_max_iterations": integer,
    "fix_standalone_glycans": boolean,
    "flash_attention_implementation": "string",
    "num_recycles": integer,
    "num_diffusion_samples": integer,
    "save_embeddings": boolean,
    "save_distogram": boolean,
    "compress_large_output_files": boolean,
    "num_seeds": integer
  }
}

{
  "deployedModelId": "string",
  "model": "string",
  "modelDisplayName": "string",
  "modelVersionId": "string",
  "predictions": [
    {
      "structure_cif": "string",
      "plddt": [
        number
      ],
      "pae": [
        [
          number
        ]
      ],
      "summary": {
        "ptm": number,
        "iptm": number,
        "fraction_disordered": number,
        "has_clash": boolean,
        "ranking_score": number,
        "chain_pair_pae_min": [
          [
            number
          ]
        ],
        "chain_pair_iptm": [
          [
            number
          ]
        ],
        "chain_ptm": [
          number
        ],
        "chain_iptm": [
          number
        ],
        "chain_ids": [
          string
        ]
      },
      "output_dir": "string"
    }
  ]
}

from google.cloud import aiplatform

PROJECT_ID = "YOUR_PROJECT_ID"
LOCATION = "us-west1"
ENDPOINT_ID = "YOUR_ENDPOINT_ID"

# Initialize AI Platform SDK
aiplatform.init(project=PROJECT_ID, location=LOCATION)

# Connect to Dedicated Endpoint
endpoint = aiplatform.Endpoint(ENDPOINT_ID)

# Define prediction payload
instances = [
    {
        "name": "preflight_check",
        "dialect": "alphafold3",
        "version": 4,
        "modelSeeds": [1],
        "sequences": [
            {
                "protein": {
                    "id": "A",
                    "sequence": "PVLSCGEWQL",
                }
            }
        ],
    }
]

parameters = {
    "dry_run": True,
}

# Execute prediction request
response = endpoint.predict(
    instances=instances,
    parameters=parameters
)
print(response.predictions)

from google.cloud import aiplatform

# Configuration
PROJECT_ID = "YOUR_PROJECT_ID"
LOCATION = "us-west1"
ENDPOINT_ID = "YOUR_ENDPOINT_ID"
STORAGE_OUTPUT_DIR = "gs://YOUR_BUCKET_NAME/alphafold_output/"

# Initialize AI Platform SDK
aiplatform.init(project=PROJECT_ID, location=LOCATION)

# Instantiate Endpoint reference
endpoint = aiplatform.Endpoint(ENDPOINT_ID)

# Define Prediction Payload
instances = [
    {
        "name": "e2e_protein_ligand_complex",
        "dialect": "alphafold3",
        "version": 4,
        "modelSeeds": [1],
        "sequences": [
            {
                "protein": {
                    "id": "A",
                    "sequence": "PVLSCGEWQL",
                    "modifications": [
                        {"ptmType": "HY3", "ptmPosition": 1}
                    ],
                }
            },
            {
                "ligand": {
                    "id": "B",
                    "ccdCodes": ["MG"],
                }
            },
        ],
    }
]

parameters = {
    "output_dir": STORAGE_OUTPUT_DIR,
}

# Execute Prediction
response = endpoint.predict(
    instances=instances,
    parameters=parameters
)

print(response.predictions)

from google.cloud import aiplatform

# Configuration
PROJECT_ID = "YOUR_PROJECT_ID"
LOCATION = "us-west1"
ENDPOINT_ID = "YOUR_ENDPOINT_ID"
STORAGE_OUTPUT_DIR = "gs://YOUR_BUCKET_NAME/af3_results/inference_only"

# Initialize AI Platform SDK
aiplatform.init(project=PROJECT_ID, location=LOCATION)

# Instantiate Endpoint reference
endpoint = aiplatform.Endpoint(ENDPOINT_ID)

# Define Prediction Payload
instances = [
    {
        "name": "inference_protein_ligand_complex",
        "dialect": "alphafold3",
        "version": 4,
        "modelSeeds": [1],
        "sequences": [
            {
                "protein": {
                    "id": "A",
                    "sequence": "PVLSCGEWQL",
                    "modifications": [
                        {"ptmType": "HY3", "ptmPosition": 1}
                    ],
                    "unpairedMsaPath": "gs://YOUR_BUCKET_NAME/path/to/unpaired.a3m",
                    "pairedMsa": "",
                    "templates": [],
                }
            },
            {
                "ligand": {
                    "id": "B",
                    "ccdCodes": ["MG"],
                }
            },
        ],
    }
]

parameters = {
    "run_data_pipeline": False,
    "output_dir": STORAGE_OUTPUT_DIR,
    "force_output_dir": True,
}

# Execute Prediction
response = endpoint.predict(
    instances=instances,
    parameters=parameters
)

print(response.predictions)

from google.cloud import aiplatform

# Configuration
PROJECT_ID = "YOUR_PROJECT_ID"
LOCATION = "us-west1"
ENDPOINT_ID = "YOUR_ENDPOINT_ID"
STORAGE_OUTPUT_DIR = "gs://YOUR_BUCKET_NAME/af3_results/inference_only_gcs_job"

# Initialize AI Platform SDK
aiplatform.init(project=PROJECT_ID, location=LOCATION)

# Instantiate Endpoint reference
endpoint = aiplatform.Endpoint(ENDPOINT_ID)

# Define Prediction Payload
instances = [
    {
        "name": "inference_only_gcs_job",
        "dialect": "alphafold3",
        "version": 4,
        "modelSeeds": [1, 2, 3],
        "sequences": [
            {
                "protein": {
                    "id": "A",
                    "sequence": "PVLSCGEWQL",
                    "unpairedMsa": "",
                    "pairedMsa": "",
                    "templates": [],
                }
            }
        ],
    }
]

parameters = {
    "output_dir": STORAGE_OUTPUT_DIR,
    "run_data_pipeline": False,
}

# Execute Prediction
response = endpoint.predict(
    instances=instances,
    parameters=parameters,
    timeout=3600.0,
)

print(response.predictions)

from google.cloud import storage

BUCKET_NAME = "your-bucket-name"
JOB_NAME = "my_alphafold_job"
STORAGE_PREFIX = f"af3_results/my_folder/{JOB_NAME}"

# Initialize GCS client
client = storage.Client(project="your-project-id")
bucket = client.bucket(BUCKET_NAME)

# Download the top-ranked 3D structure and global ranking ledger
bucket.blob(f"{STORAGE_PREFIX}/{JOB_NAME}_model.cif").download_to_filename("model.cif")
bucket.blob(f"{STORAGE_PREFIX}/{JOB_NAME}_ranking_scores.csv").download_to_filename("ranking_scores.csv")
bucket.blob(f"{STORAGE_PREFIX}/{JOB_NAME}_summary_confidences.json").download_to_filename("summary.json")

import csv
import json

print("=== Candidate Samples Ledger (ranking_scores.csv) ===")
with open("ranking_scores.csv", "r", newline="", encoding="utf-8") as f:
    rows = sorted(
        csv.DictReader(f), key=lambda x: float(x["ranking_score"]), reverse=True
    )

# Calculate column widths cleanly and readably
headers = list(rows[0].keys())
widths = {}
for col in headers:
    lengths = [len(col)] + [len(r[col]) for r in rows]
    widths[col] = max(lengths)

print("  ".join(col.rjust(widths[col]) for col in headers))
for r in rows:
    print("  ".join(r[col].rjust(widths[col]) for col in headers))

top = rows[0]
print(
    f"\nPromoted Top Candidate: Seed {int(top['seed'])}, Sample"
    f" {int(top['sample'])} (Score: {float(top['ranking_score']):.4f})"
)

print("\n=== Top Candidate Quality Validation (summary.json) ===")
with open("summary.json", "r", encoding="utf-8") as f:
    summary = json.load(f)

clash_str = (
    "DETECTED (FAIL)" if summary.get("has_clash") else "None Detected (PASS)"
)
print("Top Candidate Metrics:")
print(f"  • Ranking Score : {summary.get('ranking_score', 'N/A')}")
print(f"  • Global pTM    : {summary.get('ptm', 'N/A')}")
print(f"  • Interface ipTM: {summary.get('iptm', 'N/A')}")
print(f"  • Steric Clash  : {clash_str}")