westlake-repl/SaProt_650M_AF2

We provide two ways to use SaProt, including through huggingface class and through the same way as in esm github. Users can choose either one to use.

Huggingface model

The following code shows how to load the model.

from transformers import EsmTokenizer, EsmForMaskedLM

model_path = "/your/path/to/SaProt_650M_AF2"
tokenizer = EsmTokenizer.from_pretrained(model_path)
model = EsmForMaskedLM.from_pretrained(model_path)

#################### Example ####################
device = "cuda"
model.to(device)

seq = "M#EvVpQpL#VyQdYaKv" # Here "#" represents lower plDDT regions (plddt < 70)
tokens = tokenizer.tokenize(seq)
print(tokens)

inputs = tokenizer(seq, return_tensors="pt")
inputs = {k: v.to(device) for k, v in inputs.items()}

outputs = model(**inputs)
print(outputs.logits.shape)

"""
['M#', 'Ev', 'Vp', 'Qp', 'L#', 'Vy', 'Qd', 'Ya', 'Kv']
torch.Size([1, 11, 446])
"""

esm model

The esm version is also stored in the same folder, named SaProt_650M_AF2.pt. We provide a function to load the model.

from utils.esm_loader import load_esm_saprot

model_path = "/your/path/to/SaProt_650M_AF2.pt"
model, alphabet = load_esm_saprot(model_path)

Predict mutational effect

We provide a function to predict the mutational effect of a protein sequence. The example below shows how to predict the mutational effect at a specific position. If using the AF2 structure, we strongly recommend that you add pLDDT mask (see below).

from model.saprot.saprot_foldseek_mutation_model import SaprotFoldseekMutationModel


config = {
    "foldseek_path": None,
    "config_path": "/your/path/to/SaProt_650M_AF2", # Note this is the directory path of SaProt, not the ".pt" file
    "load_pretrained": True,
}
model = SaprotFoldseekMutationModel(**config)
tokenizer = model.tokenizer

device = "cuda"
model.eval()
model.to(device)

seq = "M#EvVpQpL#VyQdYaKv" # Here "#" represents lower plDDT regions (plddt < 70)

# Predict the effect of mutating the 3rd amino acid to A
mut_info = "V3A"
mut_value = model.predict_mut(seq, mut_info)
print(mut_value)

# Predict all effects of mutations at 3rd position
mut_pos = 3
mut_dict = model.predict_pos_mut(seq, mut_pos)
print(mut_dict)

# Predict probabilities of all amino acids at 3rd position
mut_pos = 3
mut_dict = model.predict_pos_prob(seq, mut_pos)
print(mut_dict)

"""
0.7908501625061035

{'V3A': 0.7908501625061035, 'V3C': -0.9117952585220337, 'V3D': 2.7700226306915283, 'V3E': 2.3255627155303955, 'V3F': 0.2094242423772812, 'V3G': 2.699633836746216, 'V3H': 1.240191102027893, 'V3I': 0.10231903940439224, 'V3K': 1.804598093032837,
'V3L': 1.3324960470199585, 'V3M': -0.18938277661800385, 'V3N': 2.8249857425689697, 'V3P': 0.40185314416885376, 'V3Q': 1.8361762762069702, 'V3R': 1.1899691820144653, 'V3S': 2.2159857749938965, 'V3T': 0.8813426494598389, 'V3V': 0.0, 'V3W': 0.5853186249732971, 'V3Y': 0.17449656128883362}

{'A': 0.021275954321026802, 'C': 0.0038764977362006903, 'D': 0.15396881103515625, 'E': 0.0987202599644661, 'F': 0.011895398609340191, 'G': 0.14350374042987823, 'H': 0.03334535285830498, 'I': 0.010687196627259254, 'K': 0.058634623885154724, 'L': 0.03656982257962227, 'M': 0.00798324216157198, 'N': 0.16266827285289764, 'P': 0.014419485814869404, 'Q': 0.06051575019955635, 'R': 0.03171204403042793, 'S': 0.08847439289093018, 'T': 0.023291070014238358, 'V': 0.009647775441408157, 'W': 0.017323188483715057, 'Y': 0.011487090960144997}
"""

Get protein embeddings

If you want to generate protein embeddings, you could refer to the following code. The embeddings are the average of the hidden states of the last layer.

from model.saprot.base import SaprotBaseModel
from transformers import EsmTokenizer


config = {
    "task": "base",
    "config_path": "/your/path/to/SaProt_650M_AF2", # Note this is the directory path of SaProt, not the ".pt" file
    "load_pretrained": True,
}

model = SaprotBaseModel(**config)
tokenizer = EsmTokenizer.from_pretrained(config["config_path"])

device = "cuda"
model.to(device)

seq = "M#EvVpQpL#VyQdYaKv" # Here "#" represents lower plDDT regions (plddt < 70)
tokens = tokenizer.tokenize(seq)
print(tokens)

inputs = tokenizer(seq, return_tensors="pt")
inputs = {k: v.to(device) for k, v in inputs.items()}

embeddings = model.get_hidden_states(inputs, reduction="mean")
print(embeddings[0].shape)