Try the new DeepSeek V4 Flash today. Frontier intelligence at a fraction of the cost. Here

Biology

BioNeMo Boltz2

NVIDIA NIM for Boltz-2 structure prediction and binding-affinity prediction.

Model details

Developed by
NVIDIA
Model family
BioNeMo
Use case
biology
Variant
NIM
Hardware
H100
License
NVIDIA AI Foundation Models Community License Agreement

BioNeMo Boltz-2 is a biomolecular foundation model that jointly models complex structures and binding affinities, going beyond AlphaFold3 and Boltz-1. It's the first deep learning model to approach the accuracy of physics-based free-energy perturbation methods while running roughly 1000x faster, making practical in silico screening feasible for early-stage drug discovery. It takes molecular inputs (proteins, DNA/RNA, and ligands, defined via YAML) and predicts 3D structures along with binding affinity scores, supporting both single-molecule predictions and complex multi-molecular assemblies for applications ranging from structural biology to drug discovery.

Call BioNeMo Boltz2 on Baseten using the NIM route for structure prediction.

Input

1import os
2import requests
3
4MODEL_ID = os.environ["BASETEN_MODEL_ID"]
5BASETEN_API_KEY = os.environ["BASETEN_API_KEY"]
6BASETEN_URL = (
7    f"https://model-{MODEL_ID}.api.baseten.co/environments/production/sync"
8    "/biology/mit/boltz2/predict"
9)
10
11sequence = "GIVEQCCTSICSLYQLENYCN"
12
13payload = {
14    "polymers": [
15        {
16            "id": "A",
17            "molecule_type": "protein",
18            "sequence": sequence,
19        }
20    ],
21    "recycling_steps": 3,
22    "sampling_steps": 50,
23    "diffusion_samples": 1,
24    "step_scale": 1.638,
25    "output_format": "mmcif",
26}
27
28response = requests.post(
29    BASETEN_URL,
30    headers={
31        "Authorization": f"Bearer {BASETEN_API_KEY}",
32        "Content-Type": "application/json",
33    },
34    json=payload,
35    timeout=300,
36)
37response.raise_for_status()
38result = response.json()
39print(result)
40

JSON output

1{
2    "structures": [
3        {
4            "structure": "<mmcif text>",
5            "format": "mmcif"
6        }
7    ],
8    "confidence_scores": [
9        0.87
10    ],
11    "metrics": {}
12}