Skip to content

Latest commit

 

History

History
339 lines (248 loc) · 12.4 KB

README.md

File metadata and controls

339 lines (248 loc) · 12.4 KB

Multi-Modality

AlphaFold3

Implementation of Alpha Fold 3 from the paper: "Accurate structure prediction of biomolecular interactions with AlphaFold3" in PyTorch

install

$ pip install alphafold3

Input Tensor Size Example

import torch

# Define the batch size, number of nodes, and number of features
batch_size = 1
num_nodes = 5
num_features = 64

# Generate random pair representations using torch.randn
# Shape: (batch_size, num_nodes, num_nodes, num_features)
pair_representations = torch.randn(
    batch_size, num_nodes, num_nodes, num_features
)

# Generate random single representations using torch.randn
# Shape: (batch_size, num_nodes, num_features)
single_representations = torch.randn(
    batch_size, num_nodes, num_features
)

Genetic Diffusion

Need review but basically it operates on atomic coordinates.

import torch
from alphafold3.diffusion import GeneticDiffusion

# Create an instance of the GeneticDiffusionModuleBlock
model = GeneticDiffusion(channels=3, training=True)

# Generate random input coordinates
input_coords = torch.randn(10, 100, 100, 3)

# Generate random ground truth coordinates
ground_truth = torch.randn(10, 100, 100, 3)

# Pass the input coordinates and ground truth coordinates through the model
output_coords, loss = model(input_coords, ground_truth)

# Print the output coordinates
print(output_coords)

# Print the loss value
print(loss)

Full Model Example Forward pass

import torch 
from alphafold3 import AlphaFold3

# Create random tensors
x = torch.randn(1, 5, 5, 64)  # Shape: (batch_size, seq_len, seq_len, dim)
y = torch.randn(1, 5, 64)  # Shape: (batch_size, seq_len, dim)

# Initialize AlphaFold3 model
model = AlphaFold3(
    dim=64,
    seq_len=5,
    heads=8,
    dim_head=64,
    attn_dropout=0.0,
    ff_dropout=0.0,
    global_column_attn=False,
    pair_former_depth=48,
    num_diffusion_steps=1000,
    diffusion_depth=30,
)

# Forward pass through the model
output = model(x, y)

# Print the shape of the output tensor
print(output.shape)

Docker

A basic PyTorch image is provided that includes the dependencies to run this code.

## Build the image
docker build -t af3 .

## Run the image (with GPUs)
docker run  --gpus all -it af3

Citation

@article{Abramson2024-fj,
  title    = "Accurate structure prediction of biomolecular interactions with
              {AlphaFold} 3",
  author   = "Abramson, Josh and Adler, Jonas and Dunger, Jack and Evans,
              Richard and Green, Tim and Pritzel, Alexander and Ronneberger,
              Olaf and Willmore, Lindsay and Ballard, Andrew J and Bambrick,
              Joshua and Bodenstein, Sebastian W and Evans, David A and Hung,
              Chia-Chun and O'Neill, Michael and Reiman, David and
              Tunyasuvunakool, Kathryn and Wu, Zachary and {\v Z}emgulyt{\.e},
              Akvil{\.e} and Arvaniti, Eirini and Beattie, Charles and
              Bertolli, Ottavia and Bridgland, Alex and Cherepanov, Alexey and
              Congreve, Miles and Cowen-Rivers, Alexander I and Cowie, Andrew
              and Figurnov, Michael and Fuchs, Fabian B and Gladman, Hannah and
              Jain, Rishub and Khan, Yousuf A and Low, Caroline M R and Perlin,
              Kuba and Potapenko, Anna and Savy, Pascal and Singh, Sukhdeep and
              Stecula, Adrian and Thillaisundaram, Ashok and Tong, Catherine
              and Yakneen, Sergei and Zhong, Ellen D and Zielinski, Michal and
              {\v Z}{\'\i}dek, Augustin and Bapst, Victor and Kohli, Pushmeet
              and Jaderberg, Max and Hassabis, Demis and Jumper, John M",
  journal  = "Nature",
  month    =  may,
  year     =  2024
}

Notes

-> pairwise representation -> explicit atomic positions

-> within the trunk, msa processing is de emphasized with a simpler MSA block, 4 blocks

-> msa processing -> pair weighted averaging

-> pairformer: replaces evoformer, operates on pair representation and single representation

-> pairformer 48 blocks

-> pair and single representation together with the input representation are passed to the diffusion module

-> diffusion takes in 3 tensors [pair, single representation, with new pairformer representation]

-> diffusion module operates directory on raw atom coordinates

-> standard diffusion approach, model is trained to receiev noised atomic coordinates then predict the true coordinates

-> the network learns protein structure at a variety of length scales where the denoising task at small noise emphasizes large scale structure of the system.

-> at inference time, random noise is sampled and then recurrently denoised to produce a final structure

-> diffusion module produces a distribution of answers

-> for each answer the local structure will be sharply defined

-> diffusion models are prone to hallucination where the model may hallucinate plausible looking structures

-> to counteract hallucination, they use a novel cross distillation method where they enrich the training data with alphafold multimer v2.3 predicted strutctures.

-> confidence measures predicts the atom level and pairwise errors in final structures, this is done by regressing the error in the outut of the structure mdule in training,

-> Utilizes diffusion rollout procedure for the full structure generation during training ( using a larger step suze than normal)

-> diffused predicted structure is used to permute the ground truth and ligands to compute metrics to train the confidence head.

-> confidence head uses the pairwise representation to predict the lddt (pddt) and a predicted aligned error matrix as used in alphafold 2 as well as distance error matrix which is the error in the distance matrix of the predicted structure as compared to the true structure

-> confidence measures also preduct atom level and pairwise errors

-> early stopping using a weighted average of all above metic

-> af3 can predict srtructures from input polymer sequences, rediue modifications, ligand smiles

-> uses structures below 1000 residues

-> alphafold3 is able to predict protein nuclear structures with thousnads of residues

-> Covalent modifications (bonded ligands, glycosylation, and modified protein residues and 202 nucleic acid bases) are also accurately predicted by AF

-> distills alphafold2 preductions

-> key problem in protein structure prediction is they predict static structures and not the dynamical behavior

-> multiple random seeds for either the diffusion head or network does not product an approximation of the solution ensenble

-> in future: generate large number of predictions and rank them

-> inference: top confidence sample from 5 seed runs and 5 diffusion samples per model seed for a total of 25 samples

-> interface accuracy via interface lddt which is calculated from distances netween atoms across different chains in the interface

-> uses a lddt to polymer metric which considers differences from each atom of a entity to any c or c1 polymer atom within aradius

Todo

Model Architecture

  • Implement input Embedder from Alphafold2 openfold implementation LINK

  • Implement the template module from openfold LINK

  • Implement the MSA embedding from openfold LINK

  • Fix residuals and make sure pair representation and generated output goes into the diffusion model

  • Implement reclying to fix residuals

Training pipeline

  • Get all datasets pushed to huggingface

Resources

Datasets

Smaller, start here

Benchmarks

Related Projects

Tools

Community

Books

Citations

@article{Abramson2024-fj,
  title    = "Accurate structure prediction of biomolecular interactions with
              {AlphaFold} 3",
  author   = "Abramson, Josh and Adler, Jonas and Dunger, Jack and Evans,
              Richard and Green, Tim and Pritzel, Alexander and Ronneberger,
              Olaf and Willmore, Lindsay and Ballard, Andrew J and Bambrick,
              Joshua and Bodenstein, Sebastian W and Evans, David A and Hung,
              Chia-Chun and O'Neill, Michael and Reiman, David and
              Tunyasuvunakool, Kathryn and Wu, Zachary and {\v Z}emgulyt{\.e},
              Akvil{\.e} and Arvaniti, Eirini and Beattie, Charles and
              Bertolli, Ottavia and Bridgland, Alex and Cherepanov, Alexey and
              Congreve, Miles and Cowen-Rivers, Alexander I and Cowie, Andrew
              and Figurnov, Michael and Fuchs, Fabian B and Gladman, Hannah and
              Jain, Rishub and Khan, Yousuf A and Low, Caroline M R and Perlin,
              Kuba and Potapenko, Anna and Savy, Pascal and Singh, Sukhdeep and
              Stecula, Adrian and Thillaisundaram, Ashok and Tong, Catherine
              and Yakneen, Sergei and Zhong, Ellen D and Zielinski, Michal and
              {\v Z}{\'\i}dek, Augustin and Bapst, Victor and Kohli, Pushmeet
              and Jaderberg, Max and Hassabis, Demis and Jumper, John M",
  journal  = "Nature",
  month    = "May",
  year     =  2024
}
@inproceedings{Darcet2023VisionTN,
    title   = {Vision Transformers Need Registers},
    author  = {Timoth'ee Darcet and Maxime Oquab and Julien Mairal and Piotr Bojanowski},
    year    = {2023},
    url     = {https://api.semanticscholar.org/CorpusID:263134283}
}
@article{Arora2024SimpleLA,
    title   = {Simple linear attention language models balance the recall-throughput tradeoff},
    author  = {Simran Arora and Sabri Eyuboglu and Michael Zhang and Aman Timalsina and Silas Alberti and Dylan Zinsley and James Zou and Atri Rudra and Christopher R'e},
    journal = {ArXiv},
    year    = {2024},
    volume  = {abs/2402.18668},
    url     = {https://api.semanticscholar.org/CorpusID:268063190}
}
@article{Puny2021FrameAF,
    title   = {Frame Averaging for Invariant and Equivariant Network Design},
    author  = {Omri Puny and Matan Atzmon and Heli Ben-Hamu and Edward James Smith and Ishan Misra and Aditya Grover and Yaron Lipman},
    journal = {ArXiv},
    year    = {2021},
    volume  = {abs/2110.03336},
    url     = {https://api.semanticscholar.org/CorpusID:238419638}
}
@article{Duval2023FAENetFA,
    title   = {FAENet: Frame Averaging Equivariant GNN for Materials Modeling},
    author  = {Alexandre Duval and Victor Schmidt and Alex Hernandez Garcia and Santiago Miret and Fragkiskos D. Malliaros and Yoshua Bengio and David Rolnick},
    journal = {ArXiv},
    year    = {2023},
    volume  = {abs/2305.05577},
    url     = {https://api.semanticscholar.org/CorpusID:258564608}
}
@article{Wang2022DeepNetST,
    title   = {DeepNet: Scaling Transformers to 1, 000 Layers},
    author  = {Hongyu Wang and Shuming Ma and Li Dong and Shaohan Huang and Dongdong Zhang and Furu Wei},
    journal = {ArXiv},
    year    = {2022},
    volume  = {abs/2203.00555},
    url     = {https://api.semanticscholar.org/CorpusID:247187905}
}
@inproceedings{Ainslie2023CoLT5FL,
    title   = {CoLT5: Faster Long-Range Transformers with Conditional Computation},
    author  = {Joshua Ainslie and Tao Lei and Michiel de Jong and Santiago Ontan'on and Siddhartha Brahma and Yury Zemlyanskiy and David Uthus and Mandy Guo and James Lee-Thorp and Yi Tay and Yun-Hsuan Sung and Sumit Sanghai},
    year    = {2023}
}