Skip to main content
AlphaFold 3

​
AlphaFold 3

This package provides an implementation of the inference pipeline of AlphaFold 3 - a revolutionary AI system for predicting the structure and interactions of biomolecules with unprecedented accuracy.

Quick Start

Run your first structure prediction in minutes

Installation

Complete setup guide with Docker and dependencies

Request Model Parameters

Get access to AlphaFold 3 model weights

AlphaFold Server

Use AlphaFold 3 online for non-commercial research

​
Overview

AlphaFold 3 extends the capabilities of previous versions to predict structures of:
  • Proteins with post-translational modifications
  • RNA and DNA with modified nucleotides
  • Ligands using SMILES or CCD codes
  • Protein-ligand complexes with covalent bonds
  • Biomolecular interactions across all molecule types
Any publication using this source code, model parameters, or outputs should cite the Accurate structure prediction of biomolecular interactions with AlphaFold 3 paper published in Nature.

​
Key Features

Multi-Molecule Predictions

Predict structures for proteins, RNA, DNA, ligands, and their complexes in a single model

Custom Ligands

Support for CCD codes, SMILES strings, and user-defined chemical components

Covalent Modifications

Model covalent bonds between entities and post-translational modifications

High Confidence Metrics

Per-atom pLDDT, PAE, pTM and ipTM confidence scores

Custom MSA & Templates

Provide your own multiple sequence alignments and structural templates

GPU Optimized

Efficient inference on NVIDIA A100 and H100 GPUs

​
System Requirements

AlphaFold 3 requires Linux and does not support other operating systems. Full installation requires up to 1 TB of disk space for genetic databases.

​
Minimum Hardware

  • OS: Linux (Ubuntu 22.04 LTS recommended)
  • GPU: NVIDIA GPU with Compute Capability 8.0+ (A100 80GB or H100 80GB officially supported)
  • RAM: 64 GB minimum (more for deep MSAs)
  • Storage: 1 TB SSD recommended for genetic databases
  • CUDA: Version 12.6+

​
Supported Input Sizes

  • Up to 5,120 tokens on NVIDIA A100 80 GB
  • Up to 5,120 tokens on NVIDIA H100 80 GB
  • Smaller inputs supported on GPUs with less memory

​
Input Format

AlphaFold 3 uses a flexible JSON input format that allows you to specify: 
{
  "name": "2PV7",
  "sequences": [
    {
      "protein": {
        "id": ["A", "B"],
        "sequence": "GMRESYANENQFGFKTINSDIHKIVIVGGYGKLGGLFARYLRASGYPISILDREDWAVAESILANADVVIVSVPINLTLETIERLKPYLTENMLLADLTSVKREPLAKMLEVHTGAVLGLHPMFGADIASMAKQVVVRCDGRFPERYEWLLEQIQIWGAKIYQTNATEHDHNMTYIQALRHFSTFANGLHLSKQPINLANLLALSSPIYRLELAMIGRLFAQDAELYADIIMDKSENLAVIETLKQTYDEALTFFENNDRQGFIDAFHKVRDWFGDYSEQFLKESRQLLQQANDLKQG"
      }
    }
  ],
  "modelSeeds": [1],
  "dialect": "alphafold3",
  "version": 1
}

​
Output Structure

AlphaFold 3 produces comprehensive outputs including:
  • Predicted structures in mmCIF format
  • Confidence metrics including pLDDT, PAE, pTM, and ipTM scores
  • Per-atom confidence estimates
  • Ranking scores across multiple seeds and samples
  • Optional embeddings for downstream analysis

​
Terms of Use

You may only use AlphaFold 3 model parameters if received directly from Google. Use is subject to the AlphaFold 3 Model Parameters Terms of Use.

​
Important Disclaimers

  • AlphaFold 3 and its outputs are for theoretical modeling only
  • Not intended, validated, or approved for clinical use
  • Do not use for clinical purposes or rely on for medical advice
  • Predictions have varying confidence levels - interpret carefully
  • Source code is licensed under CC-BY-NC-SA 4.0 (non-commercial use)

​
Get Started

Installation Guide

Set up AlphaFold 3 with Docker on Linux

Quick Start

Run your first prediction

Request Access

Get model parameters from Google DeepMind

​
Get in Touch

If you have any questions not covered in this documentation, please contact the AlphaFold team at [email protected]. We would love to hear your feedback and understand how AlphaFold 3 has been useful in your research.
This is not an officially supported Google product.Copyright 2024 DeepMind Technologies Limited.

Build docs developers (and LLMs) love

Get started for freeTalk to us