Try PepLink online on Hugging Face Spaces
Prefer a browser UI? Launch the hosted Gradio app in one click.

PepLink

PepLink is a Python package for peptide <-> molecular string conversion (SMILES/SELFIES).

At A Glance

Before diving into the APIs, the fastest way to understand PepLink v1 is this:

• 420 bundled unusual amino acids (non-canonical residue mappings)
• 296 bundled terminal modifications in total
• 241 N-terminal modifications
• 55 C-terminal modifications
• 3 cyclic-peptide topology classes
• 11 implemented intrachain bond chemistries

The 3 cyclic-peptide topology classes are:

• SSB: sidechain-sidechain cyclization
• SMB: sidechain-mainchain cyclization
• MMB: mainchain-mainchain cyclization, including head-to-tail macrocyclization

It currently focuses on one reliable v1 scope centered on peptide <-> SMILES/SELFIES conversion:

• aa_seqs_to_smiles(...): monomer peptide definition -> SMILES or SELFIES
• smiles_to_aa_seqs(...): standard-amino-acid SMILES or SELFIES -> peptide sequence
• list_supported_noncanonical_aas(...): inspect bundled and user-registered non-canonical amino-acid mappings
• register_noncanonical_aa(...) / register_noncanonical_aas(...): register custom non-canonical amino acids for the current Python process
• load_noncanonical_aas_from_csv(...) / register_noncanonical_aas_from_csv(...): read custom non-canonical amino acids from a user CSV file

Installation

pip install PepLink

Runtime dependencies:

• rdkit
• selfies

Quick Start

For an interactive version of the examples in this README, open examples/quick_start.ipynb.

aa_seqs_to_smiles(...)

from PepLink import aa_seqs_to_smiles

smiles = aa_seqs_to_smiles(
    "RRXXRF",
    unusual_amino_acids=[
        {"position": 3, "name": "1-NAL"},
        {"position": 4, "name": "1-NAL"},
    ],
    n_terminal="ACT",
    c_terminal="AMD",
)

print(smiles)

smiles_to_aa_seqs(...)

from PepLink import smiles_to_aa_seqs

result = smiles_to_aa_seqs("C[C@H](N)C(=O)N[C@@H](CS)C(=O)O")

print(result.sequence)            # AC
print(result.is_cyclic)           # False
print(result.cyclization)         # linear
print(result.unsupported_reason)  # None

Non-canonical amino-acid registry

from PepLink import (
    aa_seqs_to_smiles,
    list_supported_noncanonical_aas,
    register_noncanonical_aa,
)

supported = list_supported_noncanonical_aas()
print(supported["1-NAL"])

register_noncanonical_aa("MyAA", "N[C@@H](CC)C(=O)O")

smiles = aa_seqs_to_smiles(
    "AXA",
    unusual_amino_acids=[{"position": 2, "name": "MyAA"}],
)
print(smiles)

from PepLink import register_noncanonical_aas_from_csv

register_noncanonical_aas_from_csv("examples/example_custom_noncanonical_aas.csv")

Supported Scope

aa_seqs_to_smiles(...)

PepLink v1 supports monomer peptides with:

• 20 canonical amino acids plus D-forms represented by lowercase one-letter codes
• 420 bundled non-canonical amino-acid mappings
• all 241 N-terminal modifications found in all_peptides_data.json
• all 55 C-terminal modifications found in all_peptides_data.json
• 3 cyclic-peptide topology classes: SSB, SMB, and MMB
• 11 implemented intrachain bond types

Supported intrachain bond types:

• DSB
• AMD
• TIE
• DCB
• EST
• AMN
• p-XylB
• TRZB
• (E)-but-2-enyl-B
• BisMeBn-B
• but-2-ynyl-B

Meaning of the supported intrachain bond abbreviations:

Bond	Full name	Meaning
DSB	Disulfide Bond	A covalent S-S linkage between two cysteine sulfur atoms.
AMD	Amide Bond	An amide linkage formed between a carboxyl group and nitrogen; in peptides this bond has partial double-bond character, so the C-N bond is not freely rotatable.
TIE	Thioether Bond	A thioether linkage with the general form R-S-R'.
DCB	Dicarbon Bond (C=C)	A carbon-carbon double-bond crosslink.
EST	Ester Bond	An ester linkage formed from a carboxyl group and a hydroxyl group.
AMN	Amine Bond	A bond involving an amino or amine group such as -NH2, -NH-, or -N-.
p-XylB	para-Xylene thioether bridge	A para-xylene-based thioether bridge that connects two residues through sulfur atoms.
TRZB	Triazole bridge	A sidechain-sidechain linkage formed through a triazole ring bridge.
(E)-but-2-enyl-B	(E)-but-2-enyl bridge	A sidechain-sidechain crosslink bridged by an (E)-but-2-enyl group containing a C=C unit.
BisMeBn-B	Bismethylenebenzene bridge	A sidechain-sidechain crosslink bridged by a benzene ring with two methylene linkers.
but-2-ynyl-B	but-2-ynyl bridge	A sidechain-sidechain crosslink bridged by a but-2-ynyl group containing a carbon-carbon triple bond.

Common chain_participating abbreviations used in examples:

• SSB: Sidechain-Sidechain Bond
• MMB: Mainchain-Mainchain Bond
• SMB: Sidechain-Mainchain Bond

smiles_to_aa_seqs(...)

PepLink v1 intentionally keeps reverse parsing conservative.

It officially supports:

• standard amino acids only
• L/D configuration
• linear peptides
• head-to-tail cyclic peptides
• SMILES input
• SELFIES input

It does not promise reverse parsing for:

• non-canonical amino acids
• sidechain-crosslinked cyclic peptides
• terminally modified peptides
• coordination complexes

When a molecule is outside this reliable scope, smiles_to_aa_seqs(...) returns a PeptideParseResult with unsupported_reason.

Public API

aa_seqs_to_smiles(...)

aa_seqs_to_smiles(
    sequence,
    *,
    unusual_amino_acids=None,
    intrachain_bonds=None,
    n_terminal=None,
    c_terminal=None,
    output_format="smiles",
    aa_overrides=None,
    n_terminal_overrides=None,
    c_terminal_overrides=None,
) -> str

Key conventions:

• sequence uses one-letter amino-acid codes
• non-canonical residues are represented by X or x placeholders
• unusual_amino_acids must match the placeholder positions exactly
• intrachain_bonds can use either lightweight dicts or DBAASP-like nested dicts
• output_format is either "smiles" or "selfies"

Minimal direct examples:

Linear peptide

Dataset example: id=11

from PepLink import aa_seqs_to_smiles

smiles = aa_seqs_to_smiles("RVKRVWPLVIRTVIAGYNLYRAIKKK")

Single non-canonical residue

Dataset example: id=151

smiles = aa_seqs_to_smiles(
    "GIKEXKRIVQRIKDFLRNLV",
    unusual_amino_acids=[
        {"position": 5, "name": "Phg"},
    ],
)

Multiple non-canonical residues

Dataset example: id=157

smiles = aa_seqs_to_smiles(
    "GRFKRXRKKXKKLFKKIS",
    unusual_amino_acids=[
        {"position": 6, "name": "Phg"},
        {"position": 10, "name": "Phg"},
    ],
)

Terminal modifications

Dataset example: id=10360

smiles = aa_seqs_to_smiles(
    "K",
    n_terminal="C16",
    c_terminal="AMD",
)

Another real example with D-amino acids is id=8:

smiles = aa_seqs_to_smiles(
    "KVvvKWVvKvVK",
    n_terminal="C16",
    c_terminal="AMD",
)

Intrachain bond examples

Each bond type below is backed by a real record from all_peptides_data.json.

DSB

Dataset example: id=57

smiles = aa_seqs_to_smiles(
    "VTCDILSVEAKGVKLNDAACAAHCLFRGRSGGYCNGKRVCVCR",
    intrachain_bonds=[
        {"position1": 3, "position2": 34, "type": "DSB", "chain_participating": "SSB"},
        {"position1": 20, "position2": 40, "type": "DSB", "chain_participating": "SSB"},
        {"position1": 24, "position2": 42, "type": "DSB", "chain_participating": "SSB"},
    ],
)

AMD head-to-tail cyclization

Dataset example: id=105

smiles = aa_seqs_to_smiles(
    "SwFkTkSk",
    intrachain_bonds=[
        {"position1": 1, "position2": 8, "type": "AMD", "chain_participating": "MMB"},
    ],
)

TIE

Dataset example: id=1079

smiles = aa_seqs_to_smiles(
    "IXSIXLCTPGCKTGALMGCNMKTATCHCSIHVXK",
    unusual_amino_acids=[
        {"position": 2, "name": "DHB"},
        {"position": 5, "name": "DHA"},
        {"position": 33, "name": "DHA"},
    ],
    intrachain_bonds=[
        {"position1": 3, "position2": 7, "type": "TIE", "chain_participating": "SSB"},
        {"position1": 8, "position2": 11, "type": "TIE", "chain_participating": "SSB"},
        {"position1": 13, "position2": 19, "type": "TIE", "chain_participating": "SSB"},
        {"position1": 23, "position2": 26, "type": "TIE", "chain_participating": "SSB"},
        {"position1": 25, "position2": 28, "type": "TIE", "chain_participating": "SSB"},
    ],
)

DCB

Dataset example: id=4419

smiles = aa_seqs_to_smiles(
    "FLPILASLAAKFGPKLFXLVTKKX",
    unusual_amino_acids=[
        {"position": 18, "name": "AGL"},
        {"position": 24, "name": "AGL"},
    ],
    intrachain_bonds=[
        {"position1": 18, "position2": 24, "type": "DCB", "chain_participating": "SSB"},
    ],
)

EST

Dataset example: id=6917

smiles = aa_seqs_to_smiles(
    "SadAssX",
    unusual_amino_acids=[
        {"position": 7, "name": "D-Allo-Thr"},
    ],
    n_terminal="3,4-OH-4-Me-C16",
    intrachain_bonds=[
        {"position1": 0, "position2": 7, "type": "EST", "chain_participating": "MMB"},
    ],
)

AMN

Dataset example: id=19104

smiles = aa_seqs_to_smiles(
    "CANSCXYGPLTWSCXGNTK",
    unusual_amino_acids=[
        {"position": 6, "name": "DHA"},
        {"position": 15, "name": "3-OH-Asp"},
    ],
    intrachain_bonds=[
        {"position1": 1, "position2": 18, "type": "TIE", "chain_participating": "SSB"},
        {"position1": 5, "position2": 11, "type": "TIE", "chain_participating": "SSB"},
        {"position1": 4, "position2": 14, "type": "TIE", "chain_participating": "SSB"},
        {"position1": 6, "position2": 19, "type": "AMN", "chain_participating": "SSB"},
    ],
)

p-XylB

Dataset example: id=11913

smiles = aa_seqs_to_smiles(
    "cWkKkC",
    c_terminal="AMD",
    intrachain_bonds=[
        {"position1": 1, "position2": 6, "type": "p-XylB", "chain_participating": "SSB"},
    ],
)

TRZB

Dataset example: id=14660

smiles = aa_seqs_to_smiles(
    "FKXRRWQWRMKKLGAPSITXVRRAF",
    unusual_amino_acids=[
        {"position": 3, "name": "BisHomo-Pra"},
        {"position": 20, "name": "Lys(N3)"},
    ],
    intrachain_bonds=[
        {"position1": 3, "position2": 20, "type": "TRZB", "chain_participating": "SSB"},
    ],
)

(E)-but-2-enyl-B

Dataset example: id=17263

smiles = aa_seqs_to_smiles(
    "KFFKKLKKAVKKGFKKFAKV",
    intrachain_bonds=[
        {"position1": 4, "position2": 8, "type": "(E)-but-2-enyl-B", "chain_participating": "SSB"},
    ],
)

BisMeBn-B

Dataset example: id=17273

smiles = aa_seqs_to_smiles(
    "KFFKKLKKAVKKGFKKFAKV",
    intrachain_bonds=[
        {"position1": 12, "position2": 16, "type": "BisMeBn-B", "chain_participating": "SSB"},
    ],
)

but-2-ynyl-B

Dataset example: id=19191

smiles = aa_seqs_to_smiles(
    "VKRFKKFFRKFKKFV",
    c_terminal="AMD",
    intrachain_bonds=[
        {"position1": 6, "position2": 10, "type": "but-2-ynyl-B", "chain_participating": "SSB"},
    ],
)

smiles_to_aa_seqs(...)

smiles_to_aa_seqs(text, *, input_format="auto") -> PeptideParseResult

Returned fields:

• sequence
• is_cyclic
• cyclization
• normalized_smiles
• input_format
• unsupported_reason

Examples:

from PepLink import aa_seqs_to_smiles, smiles_to_aa_seqs

linear_smiles = aa_seqs_to_smiles("AC")
print(smiles_to_aa_seqs(linear_smiles))

head_to_tail_smiles = aa_seqs_to_smiles(
    "SwFkTkSk",
    intrachain_bonds=[
        {"position1": 1, "position2": 8, "type": "AMD", "chain_participating": "MMB"},
    ],
)
print(smiles_to_aa_seqs(head_to_tail_smiles))

For head-to-tail cyclic peptides, the returned sequence is normalized to a canonical rotation, because a ring has no unique start residue.

Non-canonical amino-acid registry

list_supported_noncanonical_aas(*, include_custom=True) -> dict[str, str]
load_noncanonical_aas_from_csv(csv_path) -> dict[str, str]
register_noncanonical_aa(name, smiles) -> str
register_noncanonical_aas(mapping) -> dict[str, str]
register_noncanonical_aas_from_csv(csv_path) -> dict[str, str]
clear_registered_noncanonical_aas() -> None

Key conventions:

• list_supported_noncanonical_aas(...) returns name -> SMILES mappings only for non-canonical residues
• bundled mappings contribute 420 non-canonical residue names by default
• CSV helpers expect columns name (or aa) and SMILES
• register_noncanonical_aa(...) validates and canonicalizes the input SMILES
• registered mappings are process-local and are picked up automatically by aa_seqs_to_smiles(...)
• aa_overrides is still available when you want a per-call override instead of mutating the process-wide registry

DBAASP Helper

If your source data already follows the DBAASP-style structure used in all_peptides_data.json, use from_dbaasp_record(...).

import json
from pathlib import Path

from PepLink import aa_seqs_to_smiles, from_dbaasp_record

records = json.loads(Path("all_peptides_data.json").read_text())
record = next(item for item in records if item["id"] == 57)

inputs = from_dbaasp_record(record)
smiles = aa_seqs_to_smiles(**inputs.to_api_kwargs())

Dataset Compatibility

all_peptides_data.json is the reference dataset used in this repository.

Current coverage:

• N-terminal modifications in dataset: 241 / 241 bundled
• C-terminal modifications in dataset: 55 / 55 bundled
• unusual amino-acid names in dataset: 420 / 545 bundled
• missing unusual amino-acid names: 125

Unsupported Cases

PepLink v1 intentionally rejects several categories.

• multimer peptides and interchain bonds
• coordination bonds
• reverse parsing of non-canonical / terminally modified / sidechain-crosslinked peptides
• intrachain bond types not yet implemented: ETH, CAR, IMN

Real dataset examples:

• multimer / interchain bond: id=1
• coordination bond: id=15
• unsupported bond types appear in records such as id=17389 and id=21130
• a known forward edge case that still fails in v1: id=5779

Extending Mappings

You can extend the bundled mappings without modifying PepLink source code.

Register custom unusual amino acids for the current process

from PepLink import register_noncanonical_aas

register_noncanonical_aas(
    {
        "MyAA": "N[C@@H](CC)C(=O)O",
        "MyAA2": "N[C@@H](CO)C(=O)O",
    }
)

Register custom unusual amino acids from a CSV file

Example file: examples/example_custom_noncanonical_aas.csv

from PepLink import register_noncanonical_aas_from_csv

register_noncanonical_aas_from_csv("examples/example_custom_noncanonical_aas.csv")

Add missing unusual amino acids per call

smiles = aa_seqs_to_smiles(
    "AXA",
    unusual_amino_acids=[{"position": 2, "name": "MyAA"}],
    aa_overrides={"MyAA": "N[C@@H](CC)C(=O)O"},
)

Add terminal modifications

smiles = aa_seqs_to_smiles(
    "AK",
    n_terminal="MyNCap",
    c_terminal="MyCTail",
    n_terminal_overrides={"MyNCap": "CC(=O)O"},
    c_terminal_overrides={"MyCTail": "N"},
)

Notes

• Forward SELFIES output is now implemented through the public API.
• Reverse parsing remains intentionally narrower than forward generation.
• The supported runtime implementation now lives entirely inside the PepLink/ package.
• Custom non-canonical amino-acid registrations are process-local runtime state.

Citation

If you find this project useful, please cite:

@article{leng2025predicting,
  title={Predicting and generating antibiotics against future pathogens with ApexOracle},
  author={Leng, Tianang and Wan, Fangping and Torres, Marcelo Der Torossian and de la Fuente-Nunez, Cesar},
  journal={arXiv preprint arXiv:2507.07862},
  year={2025}
}