EvoDiff
EvoDiff
Version: 1
MicrosoftLast updated August 2025

Key capabilities

About this model

EvoDiff can unconditionally sample diverse structurally-plausible proteins, generate intrinsically disordered regions, and scaffold structural motifs using only sequence information, challenging a paradigm in structure-based protein design.

Key model capabilities

Below are several use cases for EvoDiff. Currently, Azure AI Foundry supports unconditional or conditional design with EvoDiff-Seq. To use EvoDiff-MSA, we point you to our github repository for more information.
  1. Unconditional generation with EvoDiff-Seq or EvoDiff-MSA(https://github.com/microsoft/evodiff/blob/main/README.md#unconditional-generation-with-evodiff-msa )
  2. Conditional sequence generation
    1. Evolution-guided protein generation with EvoDiff-MSA
    2. Generating intrinsically disordered regions with EvoDiff-Seq and EvoDiff-MSA
    3. Scaffolding functional motifs with EvoDiff-Seq and EvoDiff-MSA

Use cases

See Responsible AI for additional considerations for responsible use.

Key use cases

Below are several use cases for EvoDiff. Currently, Azure AI Foundry supports unconditional or conditional design with EvoDiff-Seq. To use EvoDiff-MSA, we point you to our github repository for more information.

Out of scope use cases

This model is intended for use on protein sequences. It is not meant for natural language or other biological sequences, such as DNA sequences. This model will not generate sequences that are not proteins. This includes cases such as trying to generate other biological sequences, such as DNA sequences, or natural language. In other words, the model will perform best on data within the data distribution, which includes protein sequences and multiple sequence alignments (MSAs).

Pricing

Pricing is based on a number of factors, including deployment type and tokens used. See pricing details here.

Technical specs

We trained all EvoDiff sequence models on 42M sequences from UniRef50 using a dilated convolutional neural network architecture introduced in the CARP protein masked language model. We trained 38M-parameter and 640M-parameter versions for each forward corruption scheme and for left-to-right autoregressive (LRAR) decoding.

Training cut-off date

The provider has not supplied this information.

Training time

  • Hardware Type: 32GB NVIDIA V100 GPUs
  • Hours used: 4,128 (14 days per sequence model, 10 days per MSA model)
  • Cloud Provider: Azure
  • Compute Region: East US
  • Carbon Emitted: 485.21 kg

Input formats

The provider has not supplied this information.

Output formats

The provider has not supplied this information.

Supported languages

The provider has not supplied this information.

Sample JSON response

The provider has not supplied this information.

Model architecture

We investigated two types of forward processes for diffusion over discrete data modalities to determine which would be most effective. In order-agnostic autoregressive diffusion OADM , one amino acid is converted to a special mask token at each step in the forward process. After $T=L$ steps, where $L$ is the length of the sequence, the entire sequence is masked. We additionally designed discrete denoising diffusion probabilistic models D3PM for protein sequences. In EvoDiff-D3PM, the forward process corrupts sequences by sampling mutations according to a transition matrix, such that after $T$ steps the sequence is indistinguishable from a uniform sample over the amino acids. In the reverse process for both, a neural network model is trained to undo the previous corruption. The trained model can then generate new sequences starting from sequences of masked tokens or of uniformly-sampled amino acids for EvoDiff-OADM or EvoDiff-D3PM, respectively.

Long context

The provider has not supplied this information.

Optimizing model performance

The provider has not supplied this information.

Additional assets

For all other models in the EvoDiff suite, please see our github repository . We provide all generated sequences on the EvoDiff Zenodo .

Training disclosure

Training, testing and validation

We obtain sequences from the Uniref50 dataset , which contains approximately 42 million protein sequences. The Multiple Sequence Alignments (MSAs) are from the OpenFold dataset , which contains 401,381 MSAs for 140,000 unique Protein Data Bank (PDB) chains and 16,000,000 UniClust30 clusters. The intrinsically disordered regions (IDR) data was obtained from the Reverse Homology GitHub . For the scaffolding structural motif benchmark, we provide pdb and fasta files used for conditionally generating sequences in the examples/scaffolding-pdbs folder. We also provide pdb files used for conditionally generating MSAs in the examples/scaffolding-msas folder.

Distribution

Distribution channels

The provider has not supplied this information.

More information

Start using EvoDiff on Azure AI Foundry with this Jupyter Notebook . For full details, please refer to our preprint .

Responsible AI considerations

Safety techniques

The provider has not supplied this information.

Safety evaluations

The provider has not supplied this information.

Known limitations

This model will not generate sequences that are not proteins. This includes cases such as trying to generate other biological sequences, such as DNA sequences, or natural language. In other words, the model will perform best on data within the data distribution, which includes protein sequences and multiple sequence alignments (MSAs). Based on review of currently available information, EvoDiff is not be expected to provide any notable uplift in expertise to users. It is also very unlikely to create any new or add to any known CBRN or advanced autonomy risks.

Acceptable use

Acceptable use policy

This model is intended for use on protein sequences. It is not meant for natural language or other biological sequences, such as DNA sequences.

Primary Use Cases

Below are several use cases for EvoDiff. Currently, Azure AI Foundry supports unconditional or conditional design with EvoDiff-Seq. To use EvoDiff-MSA, we point you to our github repository for more information.
  1. Unconditional generation with EvoDiff-Seq or EvoDiff-MSA(https://github.com/microsoft/evodiff/blob/main/README.md#unconditional-generation-with-evodiff-msa )
  2. Conditional sequence generation
    1. Evolution-guided protein generation with EvoDiff-MSA
    2. Generating intrinsically disordered regions with EvoDiff-Seq and EvoDiff-MSA
    3. Scaffolding functional motifs with EvoDiff-Seq and EvoDiff-MSA

Out-of-Scope Use Cases

This model is intended for use on protein sequences. It is not meant for natural language or other biological sequences, such as DNA sequences.

Quality and performance evaluations

Source: Microsoft

EvoDiff-Seq Performance

The reconstruction KL (Recon KL) was calculated between the distribution of amino acids in the test set and in generated samples (n=1000). The perplexity was computed on 25k samples from the test set. The minimum Hamming distance to any train sequence of the same length (Hamming) is reported for each model as the mean ± standard deviation over the generated samples. Frechet ProtT5 distance (FPD) was calculated between the test set and generated samples. The secondary structure KL (SS KL) was calculated between the means of the predicted secondary structures of the test and generated samples.
ModelparametersRecon KLperplexityHammingFPDSS KL
Test-9.92e-41-0.003920.1011.37e-51
EvoDiff-Seq (D3PM BLOSUM)38M1.77e-217.160.83 ± 0.051.423.30e-5
EvoDiff-Seq (D3PM Uniform)38M1.48e-318.820.83 ± 0.051.313.73e-5
EvoDiff-Seq (OADM)38M1.11e-314.610.83 ± 0.070.921.61e-4
EvoDiff-Seq (D3PM BLOSUM)640M3.73e-215.740.83 ± 0.051.534.96e-4
EvoDiff-Seq (D3PM Uniform)640M2.90e-318.470.83 ± 0.051.352.13e-4
EvoDiff-Seq (OADM)640M1.26e-313.050.83 ± 0.080.881.48e-4
LRAR38M7.90e-412.380.82 ± 0.060.861.61e-4
CARP38M5.71e-125.130.74 ± 0.076.302.72e-3
LRAR640M7.01e-410.410.83 ± 0.060.631.76e-5
CARP640M3.56e-131.770.84 ± 0.051.785.03e-3
ESM-1b3650M4.91e-153.490.83 ± 0.066.675.48e-4
ESM-23650M5.00e-168.390.84 ± 0.066.793.05e-3
FoldingDiff414M5.49e-2--1.641.76e-3
RFdiffusion560M7.19e-2--1.965.98e-3
Random-1.65e-1200.85 ± 0.043.161.90e-4
Notes:
  1. Calculated between the test set and validation set.
  2. Reported value is the minimum Hamming distance between any two natural sequences of the same length in UniRef50.
  3. Due to model constraints, the maximum sequence length sampled was 1022.
  4. For the FoldingDiff baseline, 1000 structures generated by FoldingDiff were randomly selected, and the corresponding 1000 inferred sequences were inverse-folded using ESM IF. These sequences are between lengths of 50 and 128 residues.
  5. For the RFdiffusion baseline,1000 structures were generated corresponding to the UniRef train distribution length, and 1000 corresponding sequences were inverse-folded using ESM-IF.

EvoDiff-MSA performance

The perplexity is calculated based on the ability of each model to reconstruct a subsampled MSA from the validation set. "Max" and "Rand. Perplexity" indicate MaxHamming and Random subsampling, respectively, for construction of the validation MSA.
CorruptionSubsamplingParamsMaxPerplexityRand.Perplexity
EvoDiff-MSA (D3PM BLOSUM)Random100M11.358.31
EvoDiff-MSA (D3PM BLOSUM)Max100M10.987.61
EvoDiff-MSA (D3PM Uniform)Random100M10.146.77
EvoDiff-MSA (D3PM Uniform)Max100M10.066.66
EvoDiff-MSA (OADM)Random100M6.053.64
EvoDiff-MSA (OADM)Max100M6.143.60
ESM-MSA-1bMax100M11.205.89

EvoDiff-Seq structural plausibility metrics

Metrics are reported as the mean ± standard deviation for 1000 generated samples for each model.
ModelParamsESM-IF scPerplexityProteinMPNN scPerplexityOmegaFold pLDDT
Test-8.04 ± 4.043.09 ± 0.6368.25 ± 17.85
EvoDiff-Seq (D3PM BLOSUM)38M12.38 ± 2.063.80 ± 0.4942.76 ± 14.55
EvoDiff-Seq (D3PM Uniform)38M12.03 ± 2.043.77 ± 0.5042.37 ± 14.39
EvoDiff-Seq (OADM)38M11.61 ± 2.383.72 ± 0.5043.78 ± 14.18
EvoDiff-Seq (D3PM BLOSUM)640M11.86 ± 2.213.73 ± 0.4844.14 ± 13.80
EvoDiff-Seq (D3PM Uniform)640M12.29 ± 2.053.78 ± 0.4941.65 ± 14.32
EvoDiff-Seq (OADM)640M11.53 ± 2.503.71 ± 0.5244.46 ± 14.62
LRAR38M11.61 ± 2.383.64 ± 0.5648.26 ± 14.87
CARP38M9.68 ± 2.563.66 ± 0.6250.79 ± 12.06
LRAR640M10.99 ± 2.633.59 ± 0.5448.71 ± 15.47
CARP640M14.13 ± 2.424.05 ± 0.5241.56 ± 14.35
ESM-1b650M13.90 ± 2.443.47 ± 0.6858.07 ± 15.64
ESM-2650M14.02 ± 2.873.58 ± 0.6950.70 ± 15.67
Random-14.68 ± 1.973.96 ± 0.5039.97 ± 14.05

EvoDiff-MSA homolog conditioned generation

Metrics are reported as the mean ± standard deviation over 250 generated samples for each model. The first subsampling method listed describes the sampling procedure to train the model, and the second describes the subsampling procedure used for generation.
ModelscPerplexitypLDDTSeq. similarityTM score
Valid5.93 ± 3.1973.99 ± 17.8014.58 ± 21.641-
EvoDiff-MSA (OADM (Rand) - Rand MSA)9.41 ± 2.6155.99 ± 14.756.13 ± 9.880.49 ± 0.23
EvoDiff-MSA (OADM (Max) - Max MSA)9.38 ± 2.5757.08 ± 16.016.74 ± 11.000.50 ± 0.23
EvoDiff-MSA (OADM (Max) - Rand MSA)9.59 ± 2.6954.95 ± 16.836.55 ± 10.490.46 ± 0.23
ESM-MSA-1b10.05 ± 2.9251.64 ± 16.547.13 ± 11.600.40 ± 0.23
Potts10.34 ± 2.2655.46 ± 13.8212.01 ± 17.190.17 ± 0.10
Note:
  1. Sequence similarity is calculated between the original query sequence and all the sequences in the MSA.

Scaffolding performance of EvoDiff-Seq

Number of scaffolding successes out of 100 generations for RFdiffusion, EvoDiff-Seq, the LRAR baseline, the CARP baseline, and randomly sampled scaffolds (Random), for each of 17 scaffolding problems. The bottom row contains the total number of successful scaffolds generated per model.
PDBRFdiffusionEvoDiff-SeqLRARCARPRandom
1BCF10024040
6E6R7116731
2KL8880110
6EXZ420000
1YCR741312107
6VW1691000
4JHW00000
5TPN610000
4ZYP400000
3IXT252322137
7MRX70000
1PRW86870545
5IUS20000
5YUI04000
5WN900002
1QJG00000
5TRV220000
Total6101491128522

Scaffolding performance of EvoDiff-MSA

Number of scaffolding successes out of 100 generations for RFdiffusion, EvoDiff-MSA (Max), EvoDiff-MSA (Random), and the ESM-MSA baseline, for each of 17 scaffolding problems. The bottom row contains the total number of successful scaffolds generated per model.
PDBRFdiffusionEvoDiff-MSA (Max)EvoDiff-MSA (Random)ESM-MSA
1BCF1001009899
6E6R71876396
2KL888113142
6EXZ42868773
1YCR74300
6VW169434
4JHW0000
5TPN61000
4ZYP40000
3IXT25
Model Specifications
LicenseMit
Last UpdatedAugust 2025
Input TypeText
Output TypeText
ProviderMicrosoft
Languages1 Language