## 1. 获取数据

获取 GLP-1序列 uniprot GLP1_HUMAN，  
Semaglutide https://pubchem.ncbi.nlm.nih.gov/compound/56843331
同时获取Exendin4， Liraglutide， Tirzepatide 的序列

将序列整理成fasta格式。

from pathlib import Path
from io import StringIO
import requests
from Bio import SeqIO
from Bio.Seq import Seq
from Bio.SeqRecord import SeqRecord

output_dir = Path("sequence_data")
structure_dir = output_dir / "pubchem_sdf"
output_dir.mkdir(exist_ok=True)
structure_dir.mkdir(exist_ok=True)
fasta_path = output_dir / "GLP1_and_drug_peptides.fasta"


def fetch_uniprot_fasta(accession):
    url = f"https://rest.uniprot.org/uniprotkb/{accession}.fasta"
    response = requests.get(url, timeout=30)
    response.raise_for_status()
    return SeqIO.read(StringIO(response.text), "fasta")


def download_pubchem_sdf(cid):
    """Download the PubChem chemical structure because PubChem has no peptide FASTA endpoint."""
    url = f"https://pubchem.ncbi.nlm.nih.gov/rest/pug/compound/cid/{cid}/SDF"
    response = requests.get(url, timeout=60)
    response.raise_for_status()
    sdf_path = structure_dir / f"CID{cid}.sdf"
    sdf_path.write_bytes(response.content)
    return sdf_path


def fetch_pubchem_title(cid):
    url = f"https://pubchem.ncbi.nlm.nih.gov/rest/pug/compound/cid/{cid}/property/Title/JSON"
    response = requests.get(url, timeout=30)
    response.raise_for_status()
    properties = response.json()["PropertyTable"]["Properties"][0]
    return properties.get("Title", f"PubChem CID {cid}")


# PubChem stores these compounds as chemically modified molecules, not FASTA.
# The sequences below are peptide backbones; X represents Aib. Lipid and terminal
# modifications are omitted from FASTA, while the original PubChem SDF is downloaded.
peptide_data = {
    "Semaglutide|CID56843331": {
        "cid": 56843331,
        "sequence": "HXEGTFTSDVSSYLEGQAAKEFIAWLV RGRG".replace(" ", ""),
        "description": "Semaglutide backbone; N-terminal Aib=X; Lys26 lipid and terminal modifications omitted",
    },
    "Exendin-4|Exenatide|CID45588096": {
        "cid": 45588096,
        "sequence": "HGEGTFTSDLSKQMEEEAVRLFIEWLKNGGPSSGAPPPS",
        "description": "Exendin-4/Exenatide backbone; C-terminal amidation omitted",
    },
    "Liraglutide|CID16134956": {
        "cid": 16134956,
        "sequence": "HAEGTFTSDVSSYLEGQAAKEEFIAWLVRGRG",
        "description": "Liraglutide backbone; Lys26 lipid and terminal modifications omitted",
    },
    "Tirzepatide|CID156588324": {
        "cid": 156588324,
        "sequence": "YXEGTFTSDYSIXLDKIAQKAFVQWLIAGGPSSGAPPPS",
        "description": "Tirzepatide backbone; Aib=X, lipid and terminal modifications omitted",
    },
}

# GLP1_HUMAN is represented by the human GCG precursor entry P01275.
# P01116 is KRAS_HUMAN and must not be used here.
glp1_record = fetch_uniprot_fasta("P01275")
glp1_record.id = "GLP1_HUMAN|P01275"
glp1_record.description = "human proglucagon precursor containing GLP-1; UniProt GLUC_HUMAN P01275"

records = [glp1_record]
for record_id, item in peptide_data.items():
    cid = item["cid"]
    try:
        sdf_path = download_pubchem_sdf(cid)
        pubchem_title = fetch_pubchem_title(cid)
        source_note = f"PubChem SDF downloaded to {sdf_path}"
    except requests.RequestException as error:
        pubchem_title = f"PubChem CID {cid}"
        source_note = f"PubChem download unavailable: {error}"
    records.append(
        SeqRecord(
            Seq(item["sequence"]),
            id=record_id,
            description=f"{item['description']}; {pubchem_title}; {source_note}",
        )
    )

with fasta_path.open("w", encoding="utf-8") as handle:
    SeqIO.write(records, handle, "fasta")

print(f"FASTA 文件已写出: {fasta_path}")
print(f"PubChem SDF 文件目录: {structure_dir}")
for record in records:
    print(f"{record.id}: {len(record.seq)} aa\n{record.seq}")

FASTA 文件已写出: sequence_data/GLP1_and_drug_peptides.fasta
PubChem SDF 文件目录: sequence_data/pubchem_sdf
GLP1_HUMAN|P01275: 180 aa
MKSIYFVAGLFVMLVQGSWQRSLQDTEEKSRSFSASQADPLSDPDQMNEDKRHSQGTFTSDYSKYLDSRRAQDFVQWLMNTKRNRNNIAKRHDEFERHAEGTFTSDVSSYLEGQAAKEFIAWLVKGRGRRDFPEEVAIVEELGRRHADGSFSDEMNTILDNLAARDFINWLIQTKITDRK
Semaglutide|CID56843331: 31 aa
HXEGTFTSDVSSYLEGQAAKEFIAWLVRGRG
Exendin-4|Exenatide|CID45588096: 39 aa
HGEGTFTSDLSKQMEEEAVRLFIEWLKNGGPSSGAPPPS
Liraglutide|CID16134956: 32 aa
HAEGTFTSDVSSYLEGQAAKEEFIAWLVRGRG
Tirzepatide|CID156588324: 39 aa
YXEGTFTSDYSIXLDKIAQKAFVQWLIAGGPSSGAPPPS

from pathlib import Path
import shutil
import subprocess
from IPython.display import Image, display
from Bio import SeqIO
from Bio.Seq import Seq
from Bio.SeqRecord import SeqRecord

analysis_dir = Path("sequence_data")
input_fasta = analysis_dir / "GLP1_and_drug_peptides.fasta"
alignment_fasta = analysis_dir / "GLP1_drug_mafft_alignment.fasta"
logo_png = analysis_dir / "GLP1_drug_WebLogo.png"

if shutil.which("mafft") is None:
    raise RuntimeError("未找到 MAFFT。请在 biosoft 环境中安装：conda install -n biosoft -c bioconda mafft")

source_records = list(SeqIO.parse(input_fasta, "fasta"))

# P01275 is a precursor. Extract mature human GLP-1(7-37), not the 180-aa precursor.
glp1_precursor = next(record for record in source_records if record.id.startswith("GLP1_HUMAN"))
glp1_7_37_sequence = "HAEGTFTSDVSSYLEGQAAKEFIAWLVKGR"
if glp1_7_37_sequence not in str(glp1_precursor.seq):
    raise ValueError("没有在 UniProt P01275 前体中找到 GLP-1(7-37) 序列")

glp1_record = SeqRecord(
    Seq(glp1_7_37_sequence),
    id="GLP1_7-37_HUMAN",
    description="mature human GLP-1(7-37), extracted from UniProt P01275",
)

comparison_ids = [
    "Semaglutide",
    "Exendin-4",
    "Liraglutide",
    "Tirzepatide",
]
comparison_records = [glp1_record]
for name in comparison_ids:
    matching_records = [record for record in source_records if record.id.startswith(name)]
    if not matching_records:
        raise ValueError(f"输入 FASTA 中没有找到 {name}")
    record = matching_records[0]
    comparison_records.append(
        SeqRecord(Seq(str(record.seq)), id=name, description=record.description)
    )

comparison_fasta = analysis_dir / "GLP1_drug_sequences_for_alignment.fasta"
SeqIO.write(comparison_records, comparison_fasta, "fasta")

with alignment_fasta.open("w", encoding="utf-8") as output_handle:
    subprocess.run(
        ["mafft", "--auto", "--quiet", str(comparison_fasta)],
        check=True,
        stdout=output_handle,
        text=True,
    )

if shutil.which("weblogo") is None:
    raise RuntimeError("未找到 WebLogo。请在 biosoft 环境中安装：conda install -n biosoft -c bioconda weblogo")

subprocess.run(
    [
        "weblogo",
        "-f", str(alignment_fasta),
        "-o", str(logo_png),
        "-F", "png",
        "--title", "GLP-1 peptide sequence conservation",
        "--fineprint", "",
    ],
    check=True,
)

print(f"比对输入: {comparison_fasta}")
print(f"MAFFT 比对结果: {alignment_fasta}")
print(f"WebLogo 图片: {logo_png}")
display(Image(filename=str(logo_png)))

for record in SeqIO.parse(alignment_fasta, "fasta"):
    print(f"{record.id}: {record.seq}")

比对输入: sequence_data/GLP1_drug_sequences_for_alignment.fasta
MAFFT 比对结果: sequence_data/GLP1_drug_mafft_alignment.fasta
WebLogo 图片: sequence_data/GLP1_drug_WebLogo.png

GLP1_7-37_HUMAN: HAEGTFTSDVSSYLEGQAAK-EFIAWLVKGR---------
Semaglutide: HXEGTFTSDVSSYLEGQAAK-EFIAWLVRGRG--------
Exendin-4: HGEGTFTSDLSKQMEEEAVR-LFIEWLKNGGPSSGAPPPS
Liraglutide: HAEGTFTSDVSSYLEGQAAKEEFIAWLVRGRG--------
Tirzepatide: YXEGTFTSDYSIXLDKIAQK-AFVQWLIAGGPSSGAPPPS

from pathlib import Path
import pandas as pd
from IPython.display import display
from rdkit import Chem
from rdkit.Chem import Crippen, Descriptors, Lipinski, rdMolDescriptors

analysis_dir = Path("sequence_data")
structure_dir = analysis_dir / "pubchem_sdf"
properties_csv = analysis_dir / "GLP1_drug_SDF_surface_properties.csv"

sdf_files = {
    "Semaglutide": structure_dir / "CID56843331.sdf",
    "Exendin-4": structure_dir / "CID45588096.sdf",
    "Liraglutide": structure_dir / "CID16134956.sdf",
    "Tirzepatide": structure_dir / "CID156588324.sdf",
}


def read_sdf_molecule(sdf_path):
    supplier = Chem.SDMolSupplier(str(sdf_path), removeHs=False, sanitize=True)
    molecule = next((mol for mol in supplier if mol is not None), None)
    if molecule is None:
        raise ValueError(f"无法从 SDF 读取分子: {sdf_path}")
    return molecule


def sdf_property(molecule, name, default=""):
    return molecule.GetProp(name) if molecule.HasProp(name) else default


def calculate_surface_properties(name, sdf_path):
    molecule = read_sdf_molecule(sdf_path)
    Chem.rdPartialCharges.ComputeGasteigerCharges(molecule)
    partial_charges = []
    for atom in molecule.GetAtoms():
        charge = atom.GetProp("_GasteigerCharge")
        if charge not in ("nan", "-nan"):
            partial_charges.append(float(charge))

    formal_charges = [atom.GetFormalCharge() for atom in molecule.GetAtoms()]
    element_counts = {}
    for atom in molecule.GetAtoms():
        element_counts[atom.GetSymbol()] = element_counts.get(atom.GetSymbol(), 0) + 1

    # SDF/PubChem fields are metadata written in the downloaded file.
    # RDKit fields below are recalculated from the molecular graph.
    row = {
        "drug": name,
        "sdf_file": str(sdf_path),
        "molecular_formula_from_sdf": sdf_property(molecule, "PUBCHEM_MOLECULAR_FORMULA"),
        "sdf_total_charge": sdf_property(molecule, "PUBCHEM_TOTAL_CHARGE"),
        "sdf_molecular_weight_Da": sdf_property(molecule, "PUBCHEM_MOLECULAR_WEIGHT"),
        "sdf_exact_mass_Da": sdf_property(molecule, "PUBCHEM_EXACT_MASS"),
        "sdf_xlogp3": sdf_property(molecule, "PUBCHEM_XLOGP3_AA"),
        "sdf_tPSA_A2": sdf_property(molecule, "PUBCHEM_CACTVS_TPSA"),
        "rdkit_formal_charge_sum": sum(formal_charges),
        "rdkit_nonzero_formal_charge_atoms": sum(charge != 0 for charge in formal_charges),
        "rdkit_gasteiger_charge_min": min(partial_charges),
        "rdkit_gasteiger_charge_max": max(partial_charges),
        "rdkit_gasteiger_charge_abs_sum": sum(abs(charge) for charge in partial_charges),
        "rdkit_molecular_weight_Da": Descriptors.MolWt(molecule),
        "rdkit_exact_molecular_weight_Da": Descriptors.ExactMolWt(molecule),
        "rdkit_logP_XlogP3": Crippen.MolLogP(molecule),
        "rdkit_tPSA_A2": rdMolDescriptors.CalcTPSA(molecule),
        "hydrogen_bond_donors": Lipinski.NumHDonors(molecule),
        "hydrogen_bond_acceptors": Lipinski.NumHAcceptors(molecule),
        "heavy_atom_count": molecule.GetNumHeavyAtoms(),
        "rotatable_bonds": Lipinski.NumRotatableBonds(molecule),
        "aromatic_rings": rdMolDescriptors.CalcNumAromaticRings(molecule),
        "fraction_CSP3": rdMolDescriptors.CalcFractionCSP3(molecule),
        "molar_refractivity": Crippen.MolMR(molecule),
        "molecular_complexity": sdf_property(molecule, "PUBCHEM_CACTVS_COMPLEXITY"),
        "element_counts": "; ".join(f"{element}:{count}" for element, count in sorted(element_counts.items())),
    }
    return row


property_rows = []
for drug_name, sdf_path in sdf_files.items():
    if not sdf_path.exists():
        raise FileNotFoundError(f"找不到 SDF 文件，请先运行数据获取单元: {sdf_path}")
    property_rows.append(calculate_surface_properties(drug_name, sdf_path))

surface_properties = pd.DataFrame(property_rows).set_index("drug")
surface_properties.to_csv(properties_csv, encoding="utf-8-sig")

print(f"SDF 表面性质结果已保存: {properties_csv}")
print("说明：sdf_* 列来自 PubChem SDF 属性；rdkit_* 列由 RDKit 根据 SDF 分子图重新计算。")
display(surface_properties.round(3))

SDF 表面性质结果已保存: sequence_data/GLP1_drug_SDF_surface_properties.csv
说明：sdf_* 列来自 PubChem SDF 属性；rdkit_* 列由 RDKit 根据 SDF 分子图重新计算。

from pathlib import Path
import os
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import torch
from IPython.display import display
from transformers import AutoModel, AutoTokenizer
from umap import UMAP
from Bio import SeqIO

analysis_dir = Path("sequence_data")
input_fasta = analysis_dir / "GLP1_drug_sequences_for_alignment.fasta"
embedding_npy = analysis_dir / "GLP1_drug_ESM2_embeddings.npy"
embedding_csv = analysis_dir / "GLP1_drug_ESM2_embeddings.csv"
umap_csv = analysis_dir / "GLP1_drug_ESM2_UMAP_coordinates.csv"
umap_png = analysis_dir / "GLP1_drug_ESM2_UMAP.png"
model_name = "facebook/esm2_t6_8M_UR50D"

# Hugging Face model downloads use this local proxy.
proxy_url = "http://127.0.0.1:7897"
os.environ["HTTP_PROXY"] = proxy_url
os.environ["HTTPS_PROXY"] = proxy_url
os.environ["ALL_PROXY"] = proxy_url
os.environ["NO_PROXY"] = "127.0.0.1,localhost"

records = list(SeqIO.parse(input_fasta, "fasta"))
sequence_names = [record.id for record in records]
sequences = [str(record.seq).replace("-", "") for record in records]

print(f"加载模型: {model_name}")
print(f"模型下载代理: {proxy_url}")
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModel.from_pretrained(model_name)
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model = model.to(device).eval()

# Mean-pool residue representations while excluding ESM special tokens.
encoded = tokenizer(
    sequences,
    return_tensors="pt",
    padding=True,
    truncation=True,
    add_special_tokens=True,
)
encoded = {key: value.to(device) for key, value in encoded.items()}
with torch.no_grad():
    hidden_states = model(**encoded).last_hidden_state

residue_mask = encoded["attention_mask"][:, 1:-1].unsqueeze(-1).float()
residue_states = hidden_states[:, 1:-1, :]
embeddings = (residue_states * residue_mask).sum(dim=1) / residue_mask.sum(dim=1).clamp_min(1.0)
embeddings = embeddings.cpu().numpy()

np.save(embedding_npy, embeddings)
pd.DataFrame(embeddings, index=sequence_names).to_csv(embedding_csv, encoding="utf-8-sig")

n_samples = len(embeddings)
if n_samples < 3:
    raise ValueError("UMAP 至少需要 3 条序列")

umap_model = UMAP(
    n_neighbors=min(5, n_samples - 1),
    min_dist=0.2,
    metric="cosine",
    init="random",
    random_state=42,
)
umap_coordinates = umap_model.fit_transform(embeddings)
umap_table = pd.DataFrame(
    umap_coordinates,
    columns=["UMAP1", "UMAP2"],
    index=sequence_names,
)
umap_table.index.name = "sequence"
umap_table.to_csv(umap_csv, encoding="utf-8-sig")

fig, ax = plt.subplots(figsize=(8, 6))
colors = plt.cm.tab10(np.linspace(0, 1, n_samples))
ax.scatter(umap_table["UMAP1"], umap_table["UMAP2"], c=colors, s=120, edgecolor="black")
for sequence_name, row in umap_table.iterrows():
    ax.annotate(sequence_name, (row["UMAP1"], row["UMAP2"]), xytext=(6, 6), textcoords="offset points")
ax.set_title("ESM-2 sequence embeddings projected by UMAP")
ax.set_xlabel("UMAP 1")
ax.set_ylabel("UMAP 2")
ax.grid(alpha=0.25)
fig.tight_layout()
fig.savefig(umap_png, dpi=300, bbox_inches="tight")
plt.show()

print(f"Embedding 数组: {embedding_npy}, shape={embeddings.shape}")
print(f"Embedding 表格: {embedding_csv}")
print(f"UMAP 坐标: {umap_csv}")
print(f"UMAP 图片: {umap_png}")
display(umap_table.round(4))

加载模型: facebook/esm2_t6_8M_UR50D
模型下载代理: http://127.0.0.1:7897

Warning: You are sending unauthenticated requests to the HF Hub. Please set a HF_TOKEN to enable higher rate limits and faster downloads.
Loading weights: 100%|██████████| 102/102 [00:00<00:00, 8089.00it/s]
[transformers] EsmModel LOAD REPORT from: facebook/esm2_t6_8M_UR50D
Key                       | Status     | 
--------------------------+------------+-
lm_head.layer_norm.bias   | UNEXPECTED | 
lm_head.bias              | UNEXPECTED | 
lm_head.dense.bias        | UNEXPECTED | 
lm_head.dense.weight      | UNEXPECTED | 
lm_head.layer_norm.weight | UNEXPECTED | 
pooler.dense.weight       | MISSING    | 
pooler.dense.bias         | MISSING    | 

Notes:
- UNEXPECTED:	can be ignored when loading from different task/architecture; not ok if you expect identical arch.
- MISSING:	those params were newly initialized because missing from the checkpoint. Consider training on your downstream task.
/home/wangzg/miniconda3/envs/biosoft/lib/python3.10/site-packages/umap/umap_.py:1952: UserWarning: n_jobs value 1 overridden to 1 by setting random_state. Use no seed for parallelism.
  warn(

Embedding 数组: sequence_data/GLP1_drug_ESM2_embeddings.npy, shape=(5, 320)
Embedding 表格: sequence_data/GLP1_drug_ESM2_embeddings.csv
UMAP 坐标: sequence_data/GLP1_drug_ESM2_UMAP_coordinates.csv
UMAP 图片: sequence_data/GLP1_drug_ESM2_UMAP.png

from pathlib import Path
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
from IPython.display import display
from sklearn.decomposition import PCA
from sklearn.metrics.pairwise import cosine_similarity

analysis_dir = Path("sequence_data")
cosine_csv = analysis_dir / "GLP1_drug_ESM2_cosine_similarity.csv"
pca_csv = analysis_dir / "GLP1_drug_ESM2_PCA_coordinates.csv"
pca_png = analysis_dir / "GLP1_drug_ESM2_PCA.png"

# embeddings and sequence_names are generated in the previous ESM-2 cell.
if "embeddings" not in globals() or "sequence_names" not in globals():
    embedding_npy = analysis_dir / "GLP1_drug_ESM2_embeddings.npy"
    embedding_csv = analysis_dir / "GLP1_drug_ESM2_embeddings.csv"
    embeddings = np.load(embedding_npy)
    sequence_names = pd.read_csv(embedding_csv, index_col=0).index.tolist()

cosine_matrix = cosine_similarity(embeddings)
cosine_table = pd.DataFrame(
    cosine_matrix,
    index=sequence_names,
    columns=sequence_names,
)
cosine_table.to_csv(cosine_csv, encoding="utf-8-sig")

pca_model = PCA(n_components=2, random_state=42)
pca_coordinates = pca_model.fit_transform(embeddings)
pca_table = pd.DataFrame(
    pca_coordinates,
    columns=["PC1", "PC2"],
    index=sequence_names,
)
pca_table.index.name = "sequence"
pca_table.to_csv(pca_csv, encoding="utf-8-sig")

fig, ax = plt.subplots(figsize=(8, 6))
colors = plt.cm.tab10(np.linspace(0, 1, len(sequence_names)))
ax.scatter(pca_table["PC1"], pca_table["PC2"], c=colors, s=120, edgecolor="black")
for sequence_name, row in pca_table.iterrows():
    ax.annotate(sequence_name, (row["PC1"], row["PC2"]), xytext=(6, 6), textcoords="offset points")
ax.set_title(
    "ESM-2 embeddings projected by PCA "
    f"(PC1 {pca_model.explained_variance_ratio_[0]:.1%}, "
    f"PC2 {pca_model.explained_variance_ratio_[1]:.1%})"
)
ax.set_xlabel("PC1")
ax.set_ylabel("PC2")
ax.grid(alpha=0.25)
fig.tight_layout()
fig.savefig(pca_png, dpi=300, bbox_inches="tight")
plt.show()

print(f"余弦相似度矩阵: {cosine_csv}")
print(f"PCA 坐标: {pca_csv}")
print(f"PCA 图片: {pca_png}")
print("余弦相似度矩阵:")
display(cosine_table.round(4))
print("PCA 坐标:")
display(pca_table.round(4))

余弦相似度矩阵: sequence_data/GLP1_drug_ESM2_cosine_similarity.csv
PCA 坐标: sequence_data/GLP1_drug_ESM2_PCA_coordinates.csv
PCA 图片: sequence_data/GLP1_drug_ESM2_PCA.png
余弦相似度矩阵:

PCA 坐标:

	sdf_file	molecular_formula_from_sdf	sdf_total_charge	sdf_molecular_weight_Da	sdf_exact_mass_Da	sdf_xlogp3	sdf_tPSA_A2	rdkit_formal_charge_sum	rdkit_nonzero_formal_charge_atoms	rdkit_gasteiger_charge_min	...	rdkit_tPSA_A2	hydrogen_bond_donors	hydrogen_bond_acceptors	heavy_atom_count	rotatable_bonds	aromatic_rings	fraction_CSP3	molar_refractivity	molecular_complexity	element_counts
drug
Semaglutide	sequence_data/pubchem_sdf/CID56843331.sdf	C187H291N45O59	0	4114	4112.1187318	-5.8	1650	0	0	-0.508	...	1646.18	57	63	291	156	6	0.610	1040.275	9590	C:187; H:291; N:45; O:59
Exendin-4	sequence_data/pubchem_sdf/CID45588096.sdf	C184H282N50O60S	0	4187	4185.0306624	-21	1780	0	0	-0.481	...	1749.75	58	66	295	144	5	0.614	1041.334	10300	C:184; H:282; N:50; O:60; S:1
Liraglutide	sequence_data/pubchem_sdf/CID16134956.sdf	C172H265N43O51	0	3751	3749.9498161	-3.4	1510	0	0	-0.508	...	1513.76	54	55	266	137	6	0.593	955.167	8760	C:172; H:265; N:43; O:51
Tirzepatide	sequence_data/pubchem_sdf/CID156588324.sdf	C225H348N48O68	0	4813	4812.5315671	-6.8	1790	0	0	-0.508	...	1789.63	58	70	341	174	6	0.636	1223.004	11700	C:225; H:348; N:48; O:68

	UMAP1	UMAP2
sequence
GLP1_7-37_HUMAN	35.555801	12.2658
Semaglutide	36.603699	12.2495
Exendin-4	34.750198	11.6009
Liraglutide	35.598202	11.1073
Tirzepatide	36.754398	11.2388

	GLP1_7-37_HUMAN	Semaglutide	Exendin-4	Liraglutide	Tirzepatide
GLP1_7-37_HUMAN	1.0000	0.9949	0.9762	0.9934	0.9640
Semaglutide	0.9949	1.0000	0.9743	0.9940	0.9721
Exendin-4	0.9762	0.9743	1.0000	0.9765	0.9565
Liraglutide	0.9934	0.9940	0.9765	1.0000	0.9604
Tirzepatide	0.9640	0.9721	0.9565	0.9604	1.0000

	PC1	PC2
sequence
GLP1_7-37_HUMAN	0.3209	0.2854
Semaglutide	0.1140	0.3535
Exendin-4	0.2971	-0.8928
Liraglutide	0.4135	0.2721
Tirzepatide	-1.1454	-0.0182

Case 2：GLP-1 类多肽药物¶

2. 多重序列比对，WebLogo可视化¶

3. 基于 PubChem SDF 和 RDKit 的多肽药物分子表面性质与理化性质分析¶

4. Protein Language Model 序列嵌入与 UMAP 可视化¶