ESM2蛋白预训练模型 蛋白质、氨基酸向量表示

参考：
https://github.com/facebookresearch/esm
https://huggingface.co/facebook/esm2_t33_650M_UR50D
https://esmatlas.com/resources?action=fold

1、transformers版本使用

直接输入Fasta 氨基酸序列格式就行；第一次下载esm2_t33_650M_UR50D模型有点慢,有2个多G大

from transformers import BertTokenizer, AutoTokenizer,BertModel, AutoModel 

tokenizer_ = AutoTokenizer.from_pretrained("facebook/esm2_t33_650M_UR50D")

model_ = AutoModel.from_pretrained("facebook/esm2_t33_650M_UR50D")

outputs_ = model_(**tokenizer_("MEILCEDNISLSSIPNSLMQLGDGPRLYHNDFNSRDANTSEASNWTIDAENRTNLSCEGYLPPTCLSILHLQEKNWSALLTTVVIILTIAGNILVIMAVSLEKKLQNATNYFLMSLAIADMLLGFLVMPVSMLTILYGYRWPLPSKLCAIWIYLDVLFSTASIMHLCAISLDRYVAIQNPIHHSRFNSRTKAFLKIIAVWTISVGISMPIPVFGLQDDSKVFKEGSCLLADDNFVLIGSFVAFFIPLTIMVITYFLTIKSLQKEATLCVSDLSTRAKLASFSFLPQSSLSSEKLFQRSIHREPGSYAGRRTMQSISNEQKACKVLGIVFFLFVVMWCPFFITNIMAVICKESCNENVIGALLNVFVWIGYLSSAVNPLVYTLFNKTYRSAFSRYIQCQYKENRKPLQLILVNTIPALAYKSSQLQVGQKKNSQEDAEQTVDDCSMVTLGKQQSEENCTDNIETVNEKVSCV", return_tensors='pt'))


outputs_.keys(),outputs_

outputs_.last_hidden_state[:,0,:]  ##这是获取得到的 cls 蛋白整体代表的向量

tokenizer_对输入蛋白会头尾添加cls、eos特殊字符，占两个字符长度

2、下载代码离线运行向量表示

参考：https://github.com/facebookresearch/esm
直接clone下来代码

###下载模型
import torch
model, alphabet = torch.hub.load("facebookresearch/esm:main", "esm2_t33_650M_UR50D")

import torch
import esm

# Load ESM-2 model
model, alphabet = esm.pretrained.esm2_t33_650M_UR50D()
batch_converter = alphabet.get_batch_converter()
model.eval()  # disables dropout for deterministic results

# Prepare data (first 2 sequences from ESMStructuralSplitDataset superfamily / 4)
data = [
    ("protein1", "MKTVRQERLKSIVRILERSKEPVSGAQLAEELSVSRQVIVQDIAYLRSLGYNIVATPRGYVLAGG"),
    ("protein2", "KALTARQQEVFDLIRDHISQTGMPPTRAEIAQRLGFRSPNAAEEHLKALARKGVIEIVSGASRGIRLLQEE"),
    ("protein2 with mask","KALTARQQEVFDLIRDISQTGMPPTRAEIAQRLGFRSPNAAEEHLKALARKGVIEIVSGASRGIRLLQEE"),
    ("protein3",  "K A  I S Q"),
]
batch_labels, batch_strs, batch_tokens = batch_converter(data)

# Extract per-residue representations (on CPU)  ## 计算出每个氨基酸的向量，维度1024
with torch.no_grad():
    results = model(batch_tokens, repr_layers=[33], return_contacts=True)
token_representations = results["representations"][33]

# Generate per-sequence representations via averaging   ##蛋白质整个的向量，用所有氨基酸的向量平均，维度1024
# NOTE: token 0 is always a beginning-of-sequence token, so the first residue is token 1.
sequence_representations = []
for i, (_, seq) in enumerate(data):
    sequence_representations.append(token_representations[i, 1 : len(seq) + 1].mean(0))

# Look at the unsupervised self-attention map contact predictions
import matplotlib.pyplot as plt
for (_, seq), attention_contacts in zip(data, results["contacts"]):
    plt.matshow(attention_contacts[: len(seq), : len(seq)])
    plt.title(seq)
    plt.show()

单个蛋白

 #Prepare data (first 2 sequences from ESMStructuralSplitDataset superfamily / 4)
data = [
    ("protein111", "MEILCEDNISLSSIPNSLMQLGDGPRLYHNDFNSRDANTSEASNWTIDAENRTNLSCEGYLPPTCLSILHLQEKNWSALLTTVVIILTIAGNILVIMAVSLEKKLQNATNYFLMSLAIADMLLGFLVMPVSMLTILYGYRWPLPSKLCAIWIYLDVLFSTASIMHLCAISLDRYVAIQNPIHHSRFNSRTKAFLKIIAVWTISVGISMPIPVFGLQDDSKVFKEGSCLLADDNFVLIGSFVAFFIPLTIMVITYFLTIKSLQKEATLCVSDLSTRAKLASFSFLPQSSLSSEKLFQRSIHREPGSYAGRRTMQSISNEQKACKVLGIVFFLFVVMWCPFFITNIMAVICKESCNENVIGALLNVFVWIGYLSSAVNPLVYTLFNKTYRSAFSRYIQCQYKENRKPLQLILVNTIPALAYKSSQLQVGQKKNSQEDAEQTVDDCSMVTLGKQQSEENCTDNIETVNEKVSCV"),
   
]
batch_labels, batch_strs, batch_tokens = batch_converter(data)

# Extract per-residue representations (on CPU)
with torch.no_grad():
    results = model(batch_tokens, repr_layers=[33], return_contacts=True)
token_representations = results["representations"][33]
token_representations