chematic

日本語 | 中文

A pure-Rust cheminformatics library targeting RDKit feature parity — zero C/C++ by default.

Why does zero C/C++ matter?
RDKit.js, Indigo WASM, and OpenBabel all ship C++ code compiled via Emscripten.
That means 30–50 MB WASM binaries, complex build toolchains, and platform-specific build failures.
chematic compiles to a ~550 KB WASM bundle with a single wasm-pack build — no cmake, no clang,
no -sys crates, no build.rs C compilation anywhere in the dependency tree.
(The native-inchi feature is the only exception — it's opt-in and not needed for WASM.)

Live Demo

https://kent-tokyo.github.io/chematic/ — Interactive descriptor calculator, drug-likeness rules, fingerprint similarity, 3D viewer, and reaction schemes running entirely in your browser via WebAssembly.

Design Goals

Pure Rust, zero C/C++ FFI — guaranteed (default build)
No rdkit-sys, no openbabel-sys, no bindgen. Every algorithm — from SSSR ring
perception to ECFP fingerprints to force-field minimization — is implemented in 100% safe
Rust. The entire default dependency tree is verified FFI-free and WASM-compatible.

Optional exception: the native-inchi feature on chematic-inchi links the vendored
IUPAC InChI C library (v1.07.5) for bit-exact standard InChI/InChIKey. This requires a C
compiler but is completely opt-in — the default build stays FFI-free.

WASM-compatible and lightweight
All crates compile to wasm32-unknown-unknown without modification. The npm package
@kent-tokyo/chematic is ~550 KB versus 30–50 MB for C++ FFI alternatives.
No cmake, no emcc, no Emscripten toolchain required.

80+ WebAssembly API endpoints
The WASM layer exposes 80 functions covering descriptors, fingerprints, scaffold analysis,
stereoisomer enumeration, 3D geometry, diversity selection, and more — all callable from
JavaScript/TypeScript with full TypeScript type definitions.

Domain-specific algorithms
Rather than wrapping a generic graph library, chematic implements chemistry-specific
algorithms directly: Kekulization, Hückel aromaticity, CIP stereochemistry, SSSR ring
perception, Gasteiger charges, MaxMin/Butina diversity picking.

Reproducible and deterministic
Fingerprints use FNV-1a hashing with a fixed invariant ordering. Given the same SMILES
input, the same bits are always produced. No RNG, no platform-specific behavior.

Current Status

All phases complete + v0.3.x series (surpasses all major cheminformatics libraries): MCP server (AI agents), pKa prediction (15 SMARTS rules), ADMET profile (BBB/Caco-2/hERG/CYP3A4), IUPAC 25+ classes, WASM pKa/ADMET bindings, criterion benchmarks — 1,991 tests, all passing. Zero C/C++ dependencies by default.

Latest release: v0.3.2 (2026-06-15) — v0.3.0: MCP+pKa+ADMET | v0.3.1: WASM bindings | v0.3.2: criterion benchmarks

cargo test --workspace --lib --quiet                                          # 1,991 tests, all passing
cargo test -p chematic-inchi --features native-inchi --test standard_inchi  # +14 IUPAC-exact InChI tests

Quick Start

Installation

# Rust
cargo add chematic --git https://github.com/kent-tokyo/chematic --features "smiles,perception,chem,3d,fp"

# JavaScript/TypeScript
npm install @kent-tokyo/[email protected]

5-Minute Examples

Parse SMILES & check drug-likeness

use chematic_smiles::parse;
use chematic_chem::*;

let mol = parse("CC(=O)Oc1ccccc1C(=O)O")?;  // aspirin

println!("MW: {:.2}", molecular_weight(&mol));
println!("LogP: {:.2}", logp(&mol));
println!("TPSA: {:.2}", tpsa(&mol));

if lipinski_descriptor_pass(&mol) {
    println!("✓ Passes Lipinski's Rule of Five");
}

Detect rings & aromaticity

use chematic_perception::{find_sssr, assign_aromaticity};

let rings = find_sssr(&mol);
let aromatic = assign_aromaticity(&mol);

println!("Rings: {}", rings.ring_count());
// NEW in v0.1.32: Check for antiaromatic systems
if aromatic.has_antiaromaticity(&mol) {
    println!("⚠ Contains antiaromatic rings (unstable)");
}

Generate 3D coordinates

use chematic_3d::generate_and_minimize_constrained;

let coords_3d = generate_and_minimize_constrained(&mol);
// NEW in v0.1.32: Constraint satisfaction for better geometry

Calculate fingerprint similarity

use chematic_fp::tanimoto_ecfp4;

let benzene = parse("c1ccccc1")?;
let toluene = parse("Cc1ccccc1")?;
let sim = tanimoto_ecfp4(&benzene, &toluene)?;
println!("Similarity: {:.2}", sim);  // ~0.5

Preserve chemical metadata with CXSMILES

use chematic_smiles::parse_cxsmiles;

let cx = parse_cxsmiles("CCO |$ethanol$,atomProp:1.role.acceptor,^2:0|")?;
// cx.atom_labels: ["ethanol"]
// cx.atom_props: [(atom: 1, key: "role", value: "acceptor")]
// cx.atom_radicals: [None, 2, None]

Audit standardization with reports

use chematic_chem::{StandardizationPipeline, StandardizeOptions};

let opts = StandardizeOptions {
    largest_fragment_only: true,
    neutralize_charges: true,
    ..Default::default()
};
let pipeline = StandardizationPipeline::new(opts);
let (standardized, report) = pipeline.run(&mol);

println!("Status: {:?}", report.status);  // Unchanged | Modified | CompletedWithWarnings
for step in &report.steps {
    println!("  {}: changed={}", step.step.as_str(), step.changed);
}

Use from WASM/JavaScript

import init, { molecule_report_json, parse_cxsmiles_json } from 'chematic-wasm';

await init();

// Parse CXSMILES with metadata
const cx = JSON.parse(parse_cxsmiles_json("CCO |$ethanol$|"));
console.log(cx.atomLabels);  // ["ethanol"]

// Standardize with audit report
const report = JSON.parse(
    molecule_report_json("CC(=O)Oc1ccccc1C(=O)O")
);
console.log(`LogP: ${report.descriptors.logp}`);
console.log(`Lipinski: ${report.filters.lipinski_passes ? '✓' : '✗'}`);

Full Example (Rust)

use chematic_smiles::parse;
use chematic_perception::{find_sssr, assign_aromaticity};
use chematic_chem::*;
use chematic_3d::generate_and_minimize_dreiding;
use chematic_fp::tanimoto_ecfp4;

fn main() -> Result<(), Box<dyn std::error::Error>> {
    // Parse
    let benzene = parse("c1ccccc1")?;
    let toluene = parse("Cc1ccccc1")?;

    // Perception
    let rings = find_sssr(&benzene);
    let arom = assign_aromaticity(&benzene);
    println!("Benzene: {} rings, aromatic: {}", 
        rings.ring_count(), 
        arom.is_aromatic(&benzene));

    // Chemistry
    let mw = molecular_weight(&benzene);
    println!("Benzene MW: {:.2}", mw);

    // 3D
    let coords = generate_and_minimize_dreiding(&benzene);
    println!("3D coordinates generated");

    // Fingerprints
    let sim = tanimoto_ecfp4(&benzene, &toluene)?;
    println!("Benzene-Toluene similarity: {:.2}", sim);

    Ok(())
}

SMARTS substructure search

use chematic_smiles::parse;
use chematic_smarts::{parse_smarts, find_matches};

let mol = parse("CC(=O)Oc1ccccc1C(=O)O").unwrap(); // aspirin
let query = parse_smarts("[$(C(=O)O)]").unwrap();   // carboxylic / ester C
let matches = find_matches(&query, &mol);
println!("C(=O)O groups: {}", matches.len()); // 2

Molecular descriptors

use chematic_smiles::parse;
use chematic_chem::{molecular_weight, tpsa, logp_crippen, fsp3, qed, lipinski_passes};

let aspirin = parse("CC(=O)Oc1ccccc1C(=O)O").unwrap();
println!("MW:       {:.2}", molecular_weight(&aspirin)); // ~180.16
println!("TPSA:     {:.2}", tpsa(&aspirin));             // ~63.6
println!("LogP:     {:.2}", logp_crippen(&aspirin));     // ~1.2
println!("Fsp3:     {:.3}", fsp3(&aspirin));             // ~0.111
println!("QED:      {:.3}", qed(&aspirin));              // drug-likeness score
println!("Lipinski: {}", lipinski_passes(&aspirin));     // true

BRICS fragmentation

use chematic_smiles::parse;
use chematic_chem::brics_fragments;

let aspirin = parse("CC(=O)Oc1ccccc1C(=O)O").unwrap();
let frags = brics_fragments(&aspirin);
println!("fragments: {}", frags.len()); // ≥ 2

Fingerprints

use chematic_smiles::parse;
use chematic_fp::{ecfp4, atom_pair_fp, torsion_fp};

let aspirin = parse("CC(=O)Oc1ccccc1C(=O)O").unwrap();
let caffeine = parse("Cn1cnc2c1c(=O)n(c(=O)n2C)C").unwrap();

let sim_ecfp4    = ecfp4(&aspirin).tanimoto(&ecfp4(&caffeine));
let sim_atompair = atom_pair_fp(&aspirin).tanimoto(&atom_pair_fp(&caffeine));
let sim_torsion  = torsion_fp(&aspirin).tanimoto(&torsion_fp(&caffeine));

2D depiction

use chematic_smiles::parse;
use chematic_depict::depict_svg;

let caffeine = parse("Cn1cnc2c1c(=O)n(c(=O)n2C)C").unwrap();
let svg = depict_svg(&caffeine);
std::fs::write("caffeine.svg", svg).unwrap();

Highlighted depiction

use std::collections::HashSet;
use chematic_smiles::parse;
use chematic_depict::depict_svg_highlighted;

let mol = parse("c1ccncc1").unwrap(); // pyridine
let n_idx = mol.atoms().find(|(_, a)| a.element.atomic_number() == 7)
               .map(|(i, _)| i).unwrap();
let svg = depict_svg_highlighted(&mol, &HashSet::from([n_idx]), &HashSet::new());

JavaScript / TypeScript (WebAssembly)

~550 KB, zero C/C++ dependencies. Drop-in for browser or Node.js.
Compare with RDKit.js at ~30 MB built via Emscripten.

npm install @kent-tokyo/chematic

import init, {
  parse_smiles, canonical_tautomer, murcko_scaffold,
  largest_fragment, neutralize_charges,
  tanimoto_ecfp4, tanimoto_ecfp6, tanimoto_maccs,
  brics_fragments_json, mcs_smiles_json,
  get_descriptors_json, sssr_rings_json,
  enumerate_stereo_isomers_json,
  sdf_to_records_json, sdf_from_records_json,
  maxmin_picks_ecfp4_json, butina_cluster_ecfp4_json,
  shape_descriptors_json, generate_3d_minimized_pdb,
} from '@kent-tokyo/chematic';

await init();

// ── Parsing & descriptors ─────────────────────────────────────────
const mol = parse_smiles('CC(=O)Oc1ccccc1C(=O)O'); // aspirin
console.log(mol.molecular_weight()); // ~180.16
console.log(mol.qed());              // drug-likeness [0,1]
console.log(mol.sa_score());         // synthetic accessibility [1,10]
console.log(mol.lipinski_passes());  // true

// All descriptors at once (JSON object)
const desc = JSON.parse(get_descriptors_json(mol));
console.log(desc.mw, desc.tpsa, desc.logP, desc.fsp3);

// ── Molecule processing ───────────────────────────────────────────
const salt = parse_smiles('CC(=O)[O-].[Na+]');
const clean = largest_fragment(salt);        // remove Na+
const neutral = neutralize_charges(clean);   // neutralize [O-]

const tautomer = canonical_tautomer(parse_smiles('Oc1cccc2ccccc12'));
const scaffold = murcko_scaffold(parse_smiles('c1ccc(CC(=O)O)cc1'));

// ── Fingerprints & similarity ─────────────────────────────────────
const caffeine = parse_smiles('Cn1cnc2c1c(=O)n(c(=O)n2C)C');
console.log(tanimoto_ecfp4(mol, caffeine));  // ECFP4 Tanimoto
console.log(tanimoto_ecfp6(mol, caffeine));  // ECFP6 Tanimoto
console.log(tanimoto_maccs(mol, caffeine));  // MACCS Tanimoto

// ── Scaffold / fragmentation / MCS ───────────────────────────────
const frags = JSON.parse(brics_fragments_json(mol));
const mcs = mcs_smiles_json('["CC(=O)O","CC(=O)N"]');

// ── Stereochemistry ───────────────────────────────────────────────
const isomers = JSON.parse(enumerate_stereo_isomers_json(parse_smiles('C(F)(Cl)Br')));
// ["[C@@H](F)(Cl)Br","[C@H](F)(Cl)Br"]

// ── 3D geometry ───────────────────────────────────────────────────
const pdb = generate_3d_minimized_pdb(mol);
const shape = JSON.parse(shape_descriptors_json(mol));
console.log(shape.pmi1, shape.npr1, shape.asphericity);

// ── Diversity selection ───────────────────────────────────────────
const library = '["CC","c1ccccc1","CCO","CCCC","c1ccncc1"]';
const picks = JSON.parse(maxmin_picks_ecfp4_json(library, 3));
const clusters = JSON.parse(butina_cluster_ecfp4_json(library, 0.4));

// ── SDF round-trip with properties ───────────────────────────────
const records = JSON.parse(sdf_to_records_json(sdfString));
// records[0].smiles, records[0].name, records[0].properties.MW

const sdf = sdf_from_records_json(
  '["CC(=O)O"]',
  '["aspirin"]',
  '["MW\t180.16\nSource\tChEMBL"]'
);

Comparison with Other Cheminformatics Libraries

Notes:

• chematic WASM binary size measured with wasm-opt optimization; RDKit.js is the official WASM build.
• † Default build only. The optional native-inchi feature adds a cc/C-compiler build dependency for the vendored IUPAC InChI C library (v1.07.5). All other crates remain FFI-free. Verified: no *-sys crates, no cc build dependencies anywhere in the default dependency tree.

Recent Development (v0.3.x Era)

v0.3.2 (2026-06-15): Criterion benchmark suite

• chematic-chem/benches/descriptor_bench.rs — 5 descriptors in 0.68 µs/mol, ADMET in 150 µs/mol
• chematic-smarts/benches/smarts_bench.rs — SMARTS compile 1.02 µs/pat, recursive match 1.66 µs/mol
• scripts/rdkit_benchmark.py — RDKit Python comparison script

v0.3.1 (2026-06-15): WASM pKa/ADMET bindings (+34 tests → 209 total)

• MolHandle.pka_acid_value(), pka_base_value(), bbb_score(), bbb_passes(), caco2_permeability(), herg_risk_score(), cyp3a4_inhibition_risk()
• predict_pka_json(smiles) → per-site pKa JSON array
• admet_profile_json(smiles) → 15-field ADMET JSON bundle
• get_descriptors_json extended with bbbScore, caco2, hergRisk, pkaAcid, pkaBase

v0.3.0 (2026-06-15): pKa prediction + ADMET + MCP server

• pKa prediction (pka.rs): 15 SMARTS rules — carboxylic acid, phenol, thiol, amines, pyridine, imidazole, guanidine
• ADMET profile (admet.rs): BBB (Clark 2000), Caco-2 (Palm 1997), hERG risk, CYP3A4 risk, full AdmetProfile struct
• MCP server (chematic-mcp): 14 AI-callable tools — first cheminformatics library with native MCP support
• IUPAC expansion: 25+ compound classes (piperidine, morpholine, piperazine, naphthalene, sulfides)
• ETKDG torsion KB: 5 → 20+ patterns (biphenyl, sulfoxide, disulfide, nitrile, enamine...)

v0.2.11 (2026-06-14): Surpassed RDKit in 3 key domains ✨

• MMFF94 7-term force field complete (Halgren 1996): Out-of-Plane bending (OOP, 117 entries) + Stretch-Bend coupling (STRE-BEN, 282 entries)
• MAP4 fingerprint (Minervini 2020): Circular SMILES shingles — not in RDKit, superior to traditional circular FPs
• SMARTS engine optimization: LRU cache (5–20× speedup) + named functional group library (20 patterns)
• 1,941 tests, zero C/C++ dependencies (default) — pure Rust, fully WASM-compatible (~550 KB bundle); optional native-inchi feature adds IUPAC-exact InChI via vendored C lib

v0.2.9–v0.2.10: MMFF94 full stack + L-BFGS optimizer + WASM bindings

• MMFF94 complete 5-term stack (Bond/Angle/Torsion/vdW/Electrostatic) + Halgren Tables IV-VII parameter tables
• L-BFGS geometry minimizer with line search (faster convergence than steepest descent)
• Force-field API: energy breakdown, torsion scanning, per-element charges, full Cartesian control
• WASM bindings: mmff94_minimize_json, torsion_scan_json, breakdown_json, gasteiger_charges_json

v0.2.0–v0.2.8: Architecture stabilization + RDKit parity push

• v0.2.0: MHFP circular shingles fix (Lowe & Sayle 2013 spec), ERG security hardening, ~90% RDKit feature parity
• v0.2.1–v0.2.5: Canonical SMILES stereo robustness, tautomer zone blocking, virtual screening, bond inference safety
• v0.2.6–v0.2.8: Deterministic fingerprinting (FNV-1a hashing), InChI stereo/charge/isotope layers, reaction patterns

v0.1.88–v0.1.100: RDKit Gap Analysis & Closure

• v0.1.88–v0.1.90: InChI stereo layers, Brenk SMARTS, reionization, group normalization
• v0.1.91–v0.1.94: True MHFP, True ERG, Path FP stereo, SA Score corpus expansion
• v0.1.95–v0.1.100: Fingerprint canonicalization, MinHash LSH indexing, IUPAC naming, MMFF94 BCI charges, Kekulization robustness

v0.1.14–v0.1.87: Core cheminformatics foundation
For detailed historical roadmap (Phases 1–16), see tasks/todo.md.

Known Limitations

Kekulization (2 / 5,000 molecules — nearly resolved)

chematic-core's Kekulé assignment uses a 4-pass strategy:

• Pass 1/2: BFS augmenting paths (ascending / descending order).
• Pass 3: Bridgehead-N exclusion — N atoms at ring junctions (aromatic degree ≥ 3)
  donate a lone pair instead of occupying a double bond; the remaining C atoms are matched
  on a bipartite subgraph. Fixes indolizine-type systems (~109 corpus cases).
• Pass 4: Edmonds' blossom algorithm (O(n²m)) for non-bipartite C aromatic subgraphs
  with odd cycles (e.g. corannulene C₂₀H₁₀). Fixes the remaining complex polycyclic cases.

On the 5,000-molecule corpus from issue #11, only 2 molecules still fail kekulization
after these fixes:

Impact: KekuleError is returned explicitly; no silent wrong output is produced.
The boron-aromatic case is a genuine edge case; [H][H] has no heavy atoms and is
rejected by the IUPAC InChI library regardless of kekulization.

Aromaticity model (Hückel vs RDKit)

chematic uses the Hückel 4n+2 rule applied independently to each SSSR ring,
while RDKit uses a more sophisticated fused-ring electron-delocalization model.
Differences are most visible in N-heterocycles (pyridone, quinolone, indolizine).

Cascade effects on a 5,000-molecule corpus (issue #12), current status:

All three metrics are now at or near RDKit parity on the 5 000-molecule benchmark.

Aromatic ring count (95.6%) improved from the original 92.6% (at issue close)
via chematic_perception::count_aromatic_rings, which supplements the SSSR with
pairwise GF(2) XOR sub-rings (augmented_ring_set) to recover small rings missed
by the SSSR algorithm (e.g. the 5-ring of indolizine hidden behind a reported 9-ring),
then removes "envelope" rings that equal the bond-symmetric-difference of two smaller
aromatic rings to prevent double-counting. The remaining 4.4% gap reflects genuine
Hückel vs RDKit model differences in condensed N-heterocycles (pyridone, quinolone).

Repository Structure

chematic/
├── Cargo.toml               workspace root
├── CHANGELOG.md             version history
├── crates/
│   ├── chematic-core/       Atom, Bond, Molecule, Element, kekulization
│   ├── chematic-smiles/     OpenSMILES parser, writer, canonical SMILES
│   ├── chematic-perception/ SSSR ring perception, Huckel aromaticity
│   ├── chematic-mol/        MOL/SDF V2000+V3000 parser and writer
│   ├── chematic-depict/     2D SVG depiction engine (CPK colors, highlighting)
│   ├── chematic-chem/       Descriptors, BRICS, QED, standardization, scaffold
│   ├── chematic-fp/         ECFP4/6, MACCS, path, AtomPair, Torsion FP
│   ├── chematic-smarts/     SMARTS parser + VF2 subgraph isomorphism, MCS
│   ├── chematic-3d/         3D coordinate generation, PDB/XYZ formats
│   ├── chematic-rxn/        Reaction SMILES parser and writer
│   └── chematic/            Umbrella crate with feature flags
└── tasks/
    ├── todo.md              full roadmap checklist (Japanese)
    └── lessons.md           development lessons learned

Development Commands

cargo build --workspace      # build all crates
cargo test --workspace       # run all tests (736)
cargo check --workspace      # type-check without building
cargo clippy --workspace     # lints

License

Licensed under either of Apache License 2.0 or MIT License, at your option.