You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Boltz-1 and Boltz-1x are state-of-the-art open-source models to predict biomolecular structures containing combinations of proteins, RNA, DNA, and other molecules. They also support modified residues, covalent ligands and glycans, as well as conditioning the prediction on specified interaction pockets or contacts.
16
+
Boltz is a family of models for biomolecular interaction prediction. Boltz-1 was the first fully opensource model to approach AlphaFold3 accuracy. Our latest work Boltz-2 is a new biomolecular foundation model that goes beyond AlphaFold3 and Boltz-1 by jointly modeling complex structures and binding affinities, a critical component towards accurate molecular design. Boltz-2 is the first deep learning model to approach the accuracy of physics-based free-energy perturbation (FEP) methods, while running 1000x faster — making accurate in silico screening practical for early-stage drug discovery.
16
17
17
-
All the code and weights are provided under MIT license, making them freely available for both academic and commercial uses. For more information about the model, see our [technical report](https://doi.org/10.1101/2024.11.19.624167). To discuss updates, tools and applications join our [Slack channel](https://join.slack.com/t/boltz-community/shared_invite/zt-34qg8uink-V1LGdRRUf3avAUVaRvv93w).
18
+
All the code and weights are provided under MIT license, making them freely available for both academic and commercial uses. For more information about the model, see the [Boltz-1](https://doi.org/10.1101/2024.11.19.624167) and [Boltz-2](https://boltz.bio/boltz2_paper.pdf) technical reports. To discuss updates, tools and applications join our [Slack channel](https://join.slack.com/t/boltz-community/shared_invite/zt-34qg8uink-V1LGdRRUf3avAUVaRvv93w).
18
19
19
20
## Installation
21
+
22
+
> Note: we recommend installing boltz in a fresh python environment
23
+
20
24
Install boltz with PyPI (recommended):
21
25
22
26
```
@@ -29,41 +33,37 @@ or directly from GitHub for daily updates:
29
33
git clone https://github.com/jwohlwend/boltz.git
30
34
cd boltz; pip install -e .
31
35
```
32
-
> Note: we recommend installing boltz in a fresh python environment
33
36
34
37
## Inference
35
38
36
-
You can run inference using Boltz-1x with:
39
+
You can run inference using Boltz with:
37
40
38
41
```
39
42
boltz predict input_path --use_msa_server
40
43
```
41
44
42
-
Boltz currently accepts three input formats:
45
+
`input_path` should point to a YAML file, or a directory of YAML files for batched processing, describing the biomolecules you want to model and the properties you want to predict (e.g. affinity). To see all available options: `boltz predict --help` and for more information on these input formats, see our [prediction instructions](docs/prediction.md). By default, the `boltz` command will run the latest version of the model.
43
46
44
-
1. Fasta file, for most use cases
47
+
## Evaluation
45
48
46
-
2. A comprehensive YAML schema, for more complex use cases
49
+
⚠️ **Coming soon: updated evaluation code for Boltz-2!**
47
50
48
-
3. A directory containing files of the above formats, for batched processing
51
+
To encourage reproducibility and facilitate comparison with other models, on top of the existing Boltz-1 evaluation pipeline, we will soon provide the evaluation scripts and structural predictions for Boltz-2, Boltz-1, Chai-1 and AlphaFold3 on our test benchmark dataset, and our affinity predictions on the FEP+ benchamark, CASP16 and our MF-PCBA test set.
49
52
50
-
To see all available options: `boltz predict --help` and for more information on these input formats, see our [prediction instructions](docs/prediction.md).
53
+
54
+
51
55
52
-
## Evaluation
53
56
54
-
To encourage reproducibility and facilitate comparison with other models, we provide the evaluation scripts and predictions for Boltz-1, Boltz-1x, Chai-1 and AlphaFold3 on our test benchmark dataset as well as CASP15. These datasets are created to contain biomolecules different from the training data and to benchmark the performance of these models we run them with the same input MSAs and same number of recycling and diffusion steps. More details on these evaluations can be found in our [evaluation instructions](docs/evaluation.md).
55
-
56
-
57
-
57
+
## Training
58
58
59
+
⚠️ **Coming soon: updated training code for Boltz-2!**
59
60
60
-
## Training
61
+
If you're interested in retraining the model, currently for Boltz-1 but soon for Boltz-2, see our [training instructions](docs/training.md).
61
62
62
-
If you're interested in retraining the model, see our [training instructions](docs/training.md).
63
63
64
64
## Contributing
65
65
66
-
We welcome external contributions and are eager to engage with the community. Connect with us on our [Slack channel](https://join.slack.com/t/boltz-community/shared_invite/zt-34qg8uink-V1LGdRRUf3avAUVaRvv93w) to discuss advancements, share insights, and foster collaboration around Boltz-1.
66
+
We welcome external contributions and are eager to engage with the community. Connect with us on our [Slack channel](https://join.slack.com/t/boltz-community/shared_invite/zt-34qg8uink-V1LGdRRUf3avAUVaRvv93w) to discuss advancements, share insights, and foster collaboration around Boltz-2.
67
67
68
68
Boltz also runs on Tenstorrent hardware thanks to a [fork](https://github.com/moritztng/tt-boltz) by Moritz Thüning.
69
69
@@ -74,9 +74,17 @@ Our model and code are released under MIT License, and can be freely used for bo
74
74
75
75
## Cite
76
76
77
-
If you use this code or the models in your research, please cite the following paper:
77
+
If you use this code or the models in your research, please cite the following papers:
78
78
79
79
```bibtex
80
+
@article{passaro2025boltz2,
81
+
author = {Passaro, Saro and Corso, Gabriele and Wohlwend, Jeremy and Reveiz, Mateo and Thaler, Stephan and Somnath, Vignesh Ram and Portnoi, Tally and Roy, Julien and Stark, Hannes and Kwabi-Addo, David and Beaini, Dominique and Jaakkola, Tommi and Barzilay, Regina},
82
+
title = {Boltz-2: Towards Accurate and Efficient Binding Affinity Prediction},
83
+
year = {2025},
84
+
doi = {},
85
+
journal = {}
86
+
}
87
+
80
88
@article{wohlwend2024boltz1,
81
89
author = {Wohlwend, Jeremy and Corso, Gabriele and Passaro, Saro and Getz, Noah and Reveiz, Mateo and Leidal, Ken and Swiderski, Wojtek and Atkinson, Liam and Portnoi, Tally and Chinn, Itamar and Silterra, Jacob and Jaakkola, Tommi and Barzilay, Regina},
82
90
title = {Boltz-1: Democratizing Biomolecular Interaction Modeling},
Copy file name to clipboardExpand all lines: docs/evaluation.md
+23-4Lines changed: 23 additions & 4 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -1,12 +1,31 @@
1
1
# Evaluation
2
2
3
+
To encourage reproducibility and facilitate comparison with other models, we will provide the evaluation scripts and structural predictions for Boltz-2, Boltz-1, Chai-1 and AlphaFold3 on our test benchmark dataset, and our affinity predictions on the FEP+ benchamark, CASP16 and our MF-PCBA test set.
4
+
5
+
6
+
7
+
8
+
9
+
## Evaluation files
10
+
⚠️ **Coming soon updated evaluation files for Boltz-2!**
11
+
12
+
13
+
## Evaluation setup
14
+
⚠️ **Coming soon updated setup for Boltz-2!**
15
+
16
+
17
+
## Evaluation script
18
+
⚠️ **Coming soon updated evaluation scripts for Boltz-2!**
19
+
20
+
# Evaluation
21
+
3
22
To encourage reproducibility and facilitate comparison with other models, we provide the evaluation scripts and predictions for Boltz-1, Chai-1, and AlphaFold3 on our test benchmark dataset as well as CASP15. These datasets are created to contain biomolecules different from the training data and to benchmark the performance of these models we run them with the same input MSAs and the same number of recycling and diffusion steps.
4
23
5
24
6
25
7
26
8
27
9
-
## Evaluation files
28
+
## (Old) Boltz-1 Evaluation files
10
29
11
30
You can download all the MSAs, input files, output files and evaluation outputs for Boltz-1, Boltz-1x, Chai-1, and AlphaFold3 from this [Google Drive folder](https://drive.google.com/file/d/1JvHlYUMINOaqPTunI9wBYrfYniKgVmxf/view?usp=sharing).
12
31
@@ -35,7 +54,7 @@ boltz_results_final/
35
54
└── results_test.csv # Summary of evaluation results for test set
36
55
```
37
56
38
-
## Evaluation setup
57
+
## (Old) Boltz-1 Evaluation setup
39
58
40
59
We evaluate the model on two datasets:
41
60
- PDB test set: 541 targets after our validation cut-off date and at most 40% sequence similarity for proteins, 80% Tanimoto for ligands.
@@ -49,6 +68,6 @@ For fair comparison we compare the models with the following setup:
49
68
- We compare our oracle and top-1 numbers among the 5 samples.
50
69
51
70
52
-
## Evaluation script
71
+
## (Old) Boltz-1 Evaluation script
53
72
54
-
We also provide the scripts we used to evaluate the models and aggregate results. The evaluations were run through [OpenStructure](https://openstructure.org/docs/2.9.0/) version 2.8.0 (it is important to use the specific version for reproducing the results). You can find these scripts at `scripts/eval/run_evals.py` and `scripts/eval/aggregate_evals.py`.
73
+
We also provide the scripts we used to evaluate the models and aggregate results. The evaluations were run through [OpenStructure](https://openstructure.org/docs/2.9.0/) version 2.8.0 (it is important to use the specific version for reproducing the results). You can find these scripts at `scripts/eval/run_evals.py` and `scripts/eval/aggregate_evals.py`.
0 commit comments