[!TIP] The MultiMolecule team has confirmed that the provided model and checkpoints are producing the same intermediate representations as the original implementation.

The team releasing ProGen2 did not write this model card for this model so this model card has been written by the MultiMolecule team.

Model Details

ProGen2 is a GPT-J-style model pre-trained on a large corpus of protein sequences in a self-supervised fashion. This means that the model was trained on the raw amino acids of protein sequences only, with an automatic process to generate inputs and labels from those sequences. Please refer to the Training Details section for more information on the training process.

Variants

multimolecule/progen2-xlarge: The ProGen2 model pre-trained on Uniref90 and BFD30 with 6.4 billion parameters.
multimolecule/progen2-large: The ProGen2 model pre-trained on Uniref90 and BFD30 with 2.7 billion parameters.
multimolecule/progen2-bfd90: The ProGen2 model pre-trained on Uniref90 and BFD90 with 2.7 billion parameters.
multimolecule/progen2-base: The ProGen2 model pre-trained on Uniref90 and BFD30 with 764 million parameters.
multimolecule/progen2-oas: The ProGen2 model pre-trained on OAS with 764 million parameters.
multimolecule/progen2-medium: The ProGen2 model pre-trained on Uniref90 and BFD30 with 764 million parameters.
multimolecule/progen2-small: The ProGen2 model pre-trained on Uniref90 and BFD30 with 151 million parameters.

Model Specification

Variants	Num Layers	Hidden Size	Num Heads	Intermediate Size	Num Parameters (M)	FLOPs (G)	MACs (G)	Max Num Tokens
ProGen2-xlarge	32	4096	16	16384	6443.66	6735.76	3367.27	1024
ProGen2-large	2560	32	10240	2517.34	2664.21	1331.45
ProGen2-bfd90
ProGen2-base	27	1536	16	6144	764.81	826.85	413.12	2048
ProGen2-oas	1024
ProGen2-medium
ProGen2-small	12	1024	4096	151.15	167.74	83.75

Usage

The model file depends on the multimolecule library. You can install it using pip:

pip install multimolecule

Direct Use

Text Generation

You can use this model directly with a pipeline for text generation:

python

import multimolecule  # you must import multimolecule to register models
from transformers import pipeline

generator = pipeline("text-generation", model="multimolecule/progen2-xlarge")
output = generator("MGHGVSRPPVVTLR", max_new_tokens=50)

Downstream Use

Extract Features

Here is how to use this model to get the features of a given sequence in PyTorch:

python

from multimolecule import ProteinTokenizer, ProGen2Model


tokenizer = ProteinTokenizer.from_pretrained("multimolecule/progen2-xlarge")
model = ProGen2Model.from_pretrained("multimolecule/progen2-xlarge")

text = "MGHGVSRPPVVTLRPAVLDDCPVLWR"
input = tokenizer(text, return_tensors="pt")

output = model(**input)

Sequence Classification / Regression

[!NOTE] This model is not fine-tuned for any specific task. You will need to fine-tune the model on a downstream task to use it for sequence classification or regression.

Here is how to use this model as backbone to fine-tune for a sequence-level task in PyTorch:

python

import torch
from multimolecule import ProteinTokenizer, ProGen2ForSequencePrediction


tokenizer = ProteinTokenizer.from_pretrained("multimolecule/progen2-xlarge")
model = ProGen2ForSequencePrediction.from_pretrained("multimolecule/progen2-xlarge")

text = "MGHGVSRPPVVTLRPAVLDDCPVLWR"
input = tokenizer(text, return_tensors="pt")
label = torch.tensor([1])

output = model(**input, labels=label)

Token Classification / Regression

[!NOTE] This model is not fine-tuned for any specific task. You will need to fine-tune the model on a downstream task to use it for token classification or regression.

Here is how to use this model as backbone to fine-tune for a residue-level task in PyTorch:

python

import torch
from multimolecule import ProteinTokenizer, ProGen2ForTokenPrediction


tokenizer = ProteinTokenizer.from_pretrained("multimolecule/progen2-xlarge")
model = ProGen2ForTokenPrediction.from_pretrained("multimolecule/progen2-xlarge")

text = "MGHGVSRPPVVTLRPAVLDDCPVLWR"
input = tokenizer(text, return_tensors="pt")
label = torch.randint(2, (len(text), ))

output = model(**input, labels=label)

Training Details

ProGen2 used Causal Language Modeling (CLM) as the pre-training objective: given a protein sequence, the model is trained to predict the next amino acid token autoregressively.

Training Data

The ProGen2 models were pre-trained on protein sequence databases:

Uniref90: A clustered version of the UniProt database at 90% sequence identity, containing approximately 135 million sequences.
BFD30: The Big Fantastic Database clustered at 30% sequence identity, approximately one-third the size of Uniref90.
BFD90: The Big Fantastic Database clustered at 90% sequence identity, approximately twice the size of Uniref90.
OAS: The Observed Antibody Space database, clustered at 85% sequence identity.

Different model variants were trained on different combinations:

progen2-small, progen2-medium, progen2-base, progen2-large, progen2-xlarge: Trained on Uniref90 and BFD30.
progen2-bfd90: Trained on Uniref90 and BFD90.
progen2-oas: Trained on the OAS database.

Training Procedure

ProGen2 used causal language modeling (CLM) as the pre-training objective.

Pre-training

The model was trained on Google TPU-v3 pods using JAX.

Batch size: 500,000 -- 1,000,000
Steps: 350,000 -- 400,000
Optimizer: Adam(β1=0.9, β2=0.999, ε=1e-8)
Learning rate: 1e-5 -- 6e-4
Learning rate scheduler: Cosine
Learning rate warm-up: 3,000 -- 10,000 steps
Weight decay: 0.1
Maximum Gradient Norm: 0.8 -- 1.0

Citation

bibtex

@ARTICLE{Nijkamp2023-jz,
  title     = "{ProGen2}: Exploring the boundaries of protein language models",
  author    = "Nijkamp, Erik and Ruffolo, Jeffrey A and Weinstein, Eli N and
               Naik, Nikhil and Madani, Ali",
  abstract  = "Attention-based models trained on protein sequences have
               demonstrated incredible success at classification and generation
               tasks relevant for artificial-intelligence-driven protein
               design. However, we lack a sufficient understanding of how very
               large-scale models and data play a role in effective protein
               model development. We introduce a suite of protein language
               models, named ProGen2, that are scaled up to 6.4B parameters and
               trained on different sequence datasets drawn from over a billion
               proteins from genomic, metagenomic, and immune repertoire
               databases. ProGen2 models show state-of-the-art performance in
               capturing the distribution of observed evolutionary sequences,
               generating novel viable sequences, and predicting protein
               fitness without additional fine-tuning. As large model sizes and
               raw numbers of protein sequences continue to become more widely
               accessible, our results suggest that a growing emphasis needs to
               be placed on the data distribution provided to a protein
               sequence model. Our models and code are open sourced for
               widespread adoption in protein engineering. A record of this
               paper's Transparent Peer Review process is included in the
               supplemental information.",
  journal   = "Cell Syst.",
  publisher = "Elsevier BV",
  volume    =  14,
  number    =  11,
  pages     = "968--978.e3",
  month     =  nov,
  year      =  2023,
  keywords  = "fitness prediction; language modeling; protein design",
  copyright = "http://www.elsevier.com/open-access/userlicense/1.0/",
  language  = "en"
}

[!NOTE] The artifacts distributed in this repository are part of the MultiMolecule project. If MultiMolecule supports your research, please cite the MultiMolecule project as follows:

bibtex

@software{chen_2024_12638419,
  author    = {Chen, Zhiyuan and Zhu, Sophia Y.},
  title     = {MultiMolecule},
  doi       = {10.5281/zenodo.12638419},
  publisher = {Zenodo},
  url       = {https://doi.org/10.5281/zenodo.12638419},
  year      = 2024,
  month     = may,
  day       = 4
}

Contact

Please use GitHub issues of MultiMolecule for any questions or comments on the model card.

Please contact the authors of the ProGen2 paper for questions or comments on the paper/model.

License

This model implementation is licensed under the GNU Affero General Public License.

For additional terms and clarifications, please refer to our License FAQ.

SPDX-License-Identifier: AGPL-3.0-or-later

multimolecule/progen2-xlarge

ProGen2

Disclaimer

Model Details

Variants

Model Specification

Links

Usage

Direct Use

Text Generation

Downstream Use

Extract Features

Sequence Classification / Regression

Token Classification / Regression

Training Details

Training Data

Training Procedure

Pre-training

Citation

Contact

License

multimolecule/
progen2-xlarge