Skip to content

RENKIN

Computer-Aided Synthesis Planning (CASP) · Pure Rust · WebAssembly · Python
Named after 錬金 (renkin) — Japanese for alchemy: just as alchemists transformed base metals into gold, RENKIN transforms target molecules back into cheap starting materials.

CI Crates.io PyPI npm License: MIT

What is RENKIN?

RENKIN is a retrosynthesis engine that automatically plans multi-step chemical syntheses by working backwards from a target molecule to commercially available starting materials. Given a target SMILES, it searches for synthetic routes using a library of retrosynthetic reaction rules.

Try It Now

→ Open Playground

Runs entirely in WebAssembly — no server, no installation.

Open In Colab

One-click Python notebook — pip install renkin + aspirin example + RDKit visualization.

pip install renkin

import renkin
result = renkin.find_routes("CC(=O)Oc1ccccc1C(=O)O", depth=5)

Key Features

Feature Details
Pure Rust Zero C/C++ dependencies — safe, fast, cross-platform
WebAssembly Runs in the browser at near-native speed
Python bindings pip install renkin — no RDKit required
31 handcrafted rules + ~5,000 extracted via --templates Ester, amide, Suzuki, Buchwald-Hartwig, Wittig, sulfonamide, and more; extended via rdchiral-extracted templates (up to 50k)
509 building blocks Common pharma starting materials pre-loaded
A* / beam search Frequency-weighted A* with beam-width control; step_cost reduced for high-frequency templates (Phase A)
Route scoring Per-step confidence, success_probability (Retro-prob), route_cost with optional --bb-prices CSV
Constraint DSL --avoid-elements Br,I --require-elements B filters routes by element profile
Forward validation renkin-forward validate verifies each retrosynthetic step by forward prediction; pipe-friendly (stdin support)
Failure diagnostics renkin-bench --failure-taxonomy classifies unsolved targets by cause (beam limit, depth limit, template gap, stock near-miss)
Cascade search Two-stage search: fast defaults → hard cases re-run at higher beam/depth
Stability testing --quietset-out exports observations for quietset cross-config stability analysis
MCP server renkin-mcp exposes find_routes, diagnose_failure, validate_route to Claude Desktop

Quick Example

import renkin

routes = renkin.find_routes(
    smiles="CC(=O)Oc1ccccc1C(=O)O",  # Aspirin
    depth=5,
    max_routes=3,
)

for route in routes["routes"]:
    print(f"Route (depth {route['depth']}):")
    for step in route["steps"]:
        print(f"  {step['target']} → {' + '.join(step['precursors'])}")
        print(f"  via {step['rule']}")

use renkin::{chem_env::ChemEnv, search::{SearchConfig, find_routes}};

let env = ChemEnv::load("data/building_blocks.smi")?;
let config = SearchConfig { max_depth: 5, ..Default::default() };
let routes = find_routes("CC(=O)Oc1ccccc1C(=O)O", &env, &config)?;

for route in &routes {
    println!("Route depth: {}", route.depth());
}

import init, { find_routes } from './pkg/renkin.js';

await init();
const result = JSON.parse(find_routes("CC(=O)Oc1ccccc1C(=O)O", 5, 3, 0));
console.log(`Found ${result.routes_found} routes`);

How It Works

Target molecule (SMILES)
        │
        ▼
  Retrosynthetic   ←── 31 built-in + ~5k extracted (--templates)
  rule application
        │
        ▼
  Precursor set    ←── Check against 509 building blocks
        │
        ▼
  A* / BFS search  ←── Beam width, depth limit
        │
        ▼
  Synthetic routes (depth, steps, precursors)

Reaction Rules

RENKIN ships 31 handcrafted graph-based rules covering common pharmaceutical bond disconnections, plus supports up to 50k rdchiral-extracted templates via --templates:

  • Acyl disconnections: ester hydrolysis, amide cleavage (graph-based), Friedel-Crafts acylation
  • Aryl C-heteroatom: Buchwald-Hartwig (C-N), Ullmann ether (C-O), SNAr, sulfonamide cleavage
  • Aryl C-halide: C-Cl, C-I, C-F disconnections (Pd-activation / SNAr retro)
  • Aryl C-C coupling: Suzuki (graph-based), Heck, Negishi, Sonogashira
  • Sulfone disconnections: diaryl sulfone cleavage (graph-based)
  • Protecting groups: Boc, Cbz deprotection (graph-based)
  • Aliphatic: reductive amination, N-/O-alkylation, Wittig, Grignard, Michael
  • Oxidation: alcohol → carbonyl

With --templates data/templates_extracted_5000.smi, RENKIN uses ~5,000 additional rdchiral-extracted templates weighted by USPTO training frequency (Phase A), achieving 78.0% on USPTO-50k (depth=5, beam=100).

Installation

pip install renkin

[dependencies]
renkin = "0.15"

npm install renkin

See Installation for details.