What is sci-form?

sci-form is a high-performance computational chemistry library written in pure Rust that provides:

3D Molecular Conformer Generation — SMILES → 3D coordinates via ETKDGv2 algorithm
Semi-Empirical Quantum Chemistry — EHT, PM3/PM3(tm), GFN0/GFN1/GFN2-xTB
Ab-initio Methods — HF-3c (D3+gCP+SRB), CISD excited states
Neural Network Potentials — ANI-2x, ANI-TM (24 elements incl. transition metals)
Molecular Properties — HOMO/LUMO gaps, charges, dipole moments, ESP grids, DOS/PDOS, NPA/NBO
Force Fields — UFF and MMFF94 energy evaluation + strain analysis
Machine Learning — LogP, solubility, Lipinski, druglikeness, Random Forest, Gradient Boosting
3D Molecular Descriptors — WHIM, RDF, GETAWAY
Spectroscopy — NMR (¹H, ¹³C shifts, J-coupling), IR (vibrational analysis), UV-Vis (sTDA)
Stereochemistry — CIP priorities, R/S, E/Z, helical chirality (M/P), atropisomeric axes
Molecular Alignment — Kabsch SVD + quaternion optimal superposition (RMSD)
Solvation — Non-polar SASA, Generalized Born (HCT) electrostatic
Materials — Unit cells, 230 ITC space groups, framework assembly, geometry optimization with PBC
Periodic Systems — PBC-aware bonding, hapticity detection (metallocenes)
Reaction Transforms — SMIRKS atom-mapped reactant→product patterns
Fingerprints & Clustering — ECFP/Morgan, Tanimoto, Butina RMSD clustering
Reaction Path Backends — alpha GSM backend planning and TS search across UFF, MMFF94, PM3, xTB/GFN, HF-3c, and feature-gated ReaxFF

Everything is available from four entry points: Rust, Python, TypeScript/WASM, and CLI — with no C++ dependencies and native performance.

For the reaction-specific split between interpretation, path construction, and backend energetics, see Reaction Simulation Backends.

Why sci-form?

Feature	sci-form	RDKit	OpenBabel
Language	Pure Rust	C++ / Python	C++
Installation	`cargo add` / `pip install` / `npm install`	Complex build	Complex build
WASM Support	✅ Native	❌	❌
Conformer Accuracy	0.064 Å avg RMSD	—	~0.5 Å avg RMSD
Conformer Speed	60 mol/s	~50 mol/s	~30 mol/s
EHT / DOS / ESP	✅ Built-in	Partial (custom)	❌
PM3 / xTB / HF-3c	✅ Built-in	❌	❌
GFN1/GFN2-xTB	✅ Built-in	❌	❌
ANI Neural Potentials	✅ ANI-2x + ANI-TM	❌	❌
CISD Excited States	✅ Built-in	❌	❌
NMR / IR / UV-Vis	✅ Built-in	❌	❌
Transition Metals	✅ 24 elements	Limited	Limited
MMFF94	✅ Pure Rust	✅ C++ only	✅ C++ only
3D Descriptors	✅ WHIM/RDF/GETAWAY	Partial	❌
ML Models	✅ RF + GBM	❌	❌
Space Groups	✅ 230 ITC	❌	❌
Materials (MOF)	✅ Built-in	❌	Partial
Binary Size	~2 MB	~100 MB	~50 MB

🔗 The Complete Workflow

mermaid

graph LR
    A["SMILES Input"] --> B["ETKDGv2<br/>Conformer"]
    B --> C["3D Geometry<br/>+ Topology"]
    
    C --> E["Quantum Properties"]
    C --> F["ML Descriptors"]
    
    E --> E1["EHT"]
    E --> E2["PM3"]
    E --> E3["GFN-xTB"]
    
    E1 --> G["HOMO/LUMO<br/>DOS, Dipole"]
    E2 --> G
    E3 --> G
    
    C --> I["Force Fields"]
    I --> I1["UFF"]
    I --> I2["MMFF94"]
    I1 --> J["Energy"]
    I2 --> J
    
    F --> H["LogP, Solubility<br/>Lipinski, Drug-like"]
    
    G --> M["✓ Results"]
    H --> M
    J --> M

The Conformer Pipeline (ETKDGv2)

sci-form generates 3D conformers through a 9-step pipeline with automatic retry logic:

Pipeline Stages (9 Steps)

Stage	Steps	Purpose	Key Operations
Topology	1–3	Build distance constraints	SMILES → graph, 1-2/1-3/1-4 bounds, VdW effects, Floyd-Warshall smoothing
Embedding	4–6	Generate 3D coordinates	Random pick from bounds, Cayley-Menger metric, eigendecomposition, BFGS refinement
Optimization	7–9	Final refinement	ETKDG 3D force field (846 CSD torsions), UFF inversions, validation checks

Automatic Retry Loop

If any check fails, retry up to $10 N$ times (default) before falling back to random box placement:

❌ Metric matrix has non-positive eigenvalues
❌ Energy/atom after minimization > 0.05 kcal/mol
❌ Tetrahedral center volume sign mismatch
❌ Chiral inversion fails
❌ Planarity or double-bond geometry violations

→ Detailed explanation in Algorithm Overview

Quantum Chemistry Methods

sci-form v0.4.3 offers three complementary quantum chemistry approaches, each optimized for different use cases:

1️⃣ Extended Hückel Theory (EHT)

Gateway method for semi-empirical orbital properties — fast, stable, supported on all elements.

Capabilities:

Wolfsberg-Helmholtz Hamiltonian with STO-3G basis
Löwdin orthogonalization for canonical MOs
HOMO/LUMO energies, occupation numbers
Density of States (DOS) + per-atom PDOS with Gaussian smearing
Volumetric orbital grids (3D probability density)
Marching Cubes isosurface meshes for visualization

Properties computed from EHT:

Mulliken & Löwdin population analysis (per-atom charges)
Dipole moments (bond + lone-pair contributions in Debye)
Electrostatic potential (ESP) on 3D grid from Mulliken charges

Supported elements: H, B, C, N, O, F, Si, P, S, Cl, Br, I + all transition metals (Ti→Au)

→ EHT Algorithm Details

2️⃣ PM3 (NDDO Semi-Empirical SCF)

Full self-consistent field for accurate thermochemistry — accounts for all valence electron repulsion.

Capabilities:

Neglect of Diatomic Differential Overlap (NDDO) parameterization
Full SCF iteration with Fock matrix
Heat of formation (kcal/mol) — critical for reaction energetics
HOMO/LUMO energies, Mulliken charges, SCF iteration count
Convergence diagnostics (converged bool)

Supported elements: H, C, N, O, F, P, S, Cl, Br, I (14 main-group)

Typical convergence: 10–30 SCF cycles with $Δ P < 10^{- 6}$

Use case: Thermochemistry, activation barriers, charged species, reaction mechanisms

3️⃣ GFN-xTB (Generalized Tight-Binding, GFN0)

Ultra-fast with transition metal support — self-consistent charge (SCC) correction for modern accuracy.

Capabilities:

Slater-Koster tight-binding Hamiltonian
Self-consistent charge iterative corrections
Total energy = electronic + repulsive energy
HOMO/LUMO gap, Mulliken charges, SCC iteration count
Convergence diagnostics (converged bool)

Supported elements: H, B, C, N, O, F, Si, P, S, Cl, Br, I + all transition metals (Ti(22), Cr(24), Mn(25), Fe(26), Co(27), Ni(28), Cu(29), Zn(30), Ru(44), Pd(46), Ag(47), Pt(78), Au(79))

Typical convergence: 5–15 SCC cycles, ~100 ms on large molecules

Use case: Rapid screening of large sets, organometallics, cluster models, kinetic barriers

Comparison: EHT vs PM3 vs xTB

Property	EHT	PM3	GFN-xTB
Elements	H–F + TM	H, C, N, O, F, P, S, Cl, Br, I	H–I + 25 TM ✅
SCF / SCC	Fixed orbital	NDDO SCF ✅	Tight-binding SCC ✅
Heat of Formation	❌	✅ (kcal/mol)	❌ (use total E)
Speed	Fastest (~1 ms)	Medium (~1 s)	Fast (~100 ms)
Accuracy	Orbital properties	Thermochemistry	Screening + metals
Typical Use	Orbitals, DOS, ESP	Reaction barriers	Large-scale QM

Quantum Chemistry (EHT)

The Extended Hückel Theory module provides semi-empirical molecular orbital calculations based on the Wolfsberg-Helmholtz approximation:

Key Capabilities

Overlap integrals — STO-3G contracted Gaussians for H, C, N, O, F, S, Cl + all transition metals (Ti–Au)
Hamiltonian — diagonal Hückel parameters, off-diagonal Wolfsberg-Helmholtz $H_{i j} = \frac{K}{2} S_{i j} (H_{i i} + H_{j j})$
Orbital properties — HOMO/LUMO energies, density of states (DOS), volumetric orbital grids
Population analysis — Mulliken & Löwdin charges, dipole moments (Debye)
Molecular visualization — Marching Cubes isosurfaces, color-coded density meshes

From EHT, Compute:

Property	Output	Format
Orbital Energies	1D array (eV)	JSON array
HOMO/LUMO Gap	Single value (eV)	Scalar
Population	Per-atom charges	JSON array
Dipole Moment	Vector + magnitude (Debye)	[x, y, z], scalar
Density of States	Total DOS, PDOS (Gaussian smearing)	JSON curves
Orbital Grid	Volumetric MO density	3D F32 array
Mesh	Isosurface triangulation	Vertices, normals, indices

→ See EHT Algorithm for mathematical details.

Machine Learning Properties (No 3D Required)

For ultra-fast virtual screening without quantum chemistry:

Molecular Descriptors (17 features):

Molecular weight, heavy atoms, hydrogens, bonds
Rotatable bonds, HBD/HBA, FSP3
Topological: Wiener index, ring count, aromaticity, Balaban J
Electronic: Sum electronegativity, sum polarizability

Predicted Properties:

LogP — partition coefficient (Wildman-Crippen model)
Molar Refractivity — optical property (Ghose-Crippen)
Log Solubility — aqueous solubility (Delaney ESOL)
Lipinski Ro5 — violations count + pass/fail flag
Druglikeness — composite score

⚡ Speed: SMILES → properties in ~1 μs (topology only, no 3D coordinates needed)

Property Prediction Speed Comparison

Method	Speed	Elements	Accuracy	Use Case
ML Predict	~1 μs	All (SMILES)	Moderate	Early filtering
EHT	~1 ms	H–Au	Medium	Electronics
PM3	~1 s	14 main-group	High (thermochemistry)	Reaction design
GFN-xTB	~100 ms	25 (incl. TM)	High (screening + metals)	Large-scale QM

Force Fields & Molecular Mechanics

Evaluate strain energy and optimize molecular geometry:

UFF (Universal Force Field)

Coverage: 50+ element types including transition metals
Energy terms: Morse bond stretch, angle, out-of-plane, torsion, van der Waals
Use: Refinement after embedding, geometry optimization

MMFF94 (Merck Molecular Force Field)

Coverage: Main-group organic chemistry, parameterized from QM
Energy terms: Quartic bond stretch, cubic angle bend, 3-term torsion (Fourier), Halgren 14-7 vdW
Accuracy: ~0.08 Å RMSD vs experiment on small molecules

→ Force Fields Algorithm

Molecular Alignment

Superpose two molecular conformers and compute RMSD:

Kabsch algorithm — SVD-based optimal rotation, $O (N \cdot min (N, 3)^{2})$
Quaternion alignment — Coutsias 2004 4×4 eigenproblem, more numerically stable for large $N$
Output: Rotation matrix, optimal RMSD value, aligned coordinates

→ Alignment Algorithm

🌐 Multi-Platform Entry Points

Rust

Native library with full API:

rust

let conf = sci_form::embed("CCO", 42)?;
let pm3 = sci_form::compute_pm3(&conf.elements, &pos)?;
let xtb = sci_form::compute_xtb(&conf.elements, &pos)?;
let props = sci_form::predict_ml_properties(&desc)?;

Python

PyO3 bindings with NumPy integration:

python

import sci_form
conf = sci_form.embed("CCO", seed=42)
pm3 = sci_form.pm3_calculate(conf.elements, conf.coords)
xtb = sci_form.xtb_calculate(conf.elements, conf.coords)
props = sci_form.ml_predict("CCO")

TypeScript / WASM

Isomorphic bindings for browser, Node.js, Deno, Bun:

typescript

import init, { embed, compute_pm3, compute_xtb } from 'sci-form-wasm';
await init();
const result = JSON.parse(embed('CCO', 42));
const pm3 = JSON.parse(compute_pm3(...));

CLI

Command-line interface with batch processing:

bash

sci-form embed "CCO" -f xyz
sci-form batch molecules.smi -t 8 --output results.json
sci-form energy "CCO" --from-smiles

📖 Next Steps

Getting Started — Install and run your first calculation
Rust API — Core library functions and types
Python API — PyO3 bindings with examples
TypeScript API — WASM bindings and web integration
CLI Guide — Command-line usage
Algorithm Deep Dive — Mathematical foundations, EHT, PM3, xTB, force fields

What is sci-form? ​

Why sci-form? ​

🔗 The Complete Workflow ​

The Conformer Pipeline (ETKDGv2) ​

Pipeline Stages (9 Steps) ​

Automatic Retry Loop ​

Quantum Chemistry Methods ​

1️⃣ Extended Hückel Theory (EHT) ​

2️⃣ PM3 (NDDO Semi-Empirical SCF) ​

3️⃣ GFN-xTB (Generalized Tight-Binding, GFN0) ​

Comparison: EHT vs PM3 vs xTB ​

Quantum Chemistry (EHT) ​

Key Capabilities ​

From EHT, Compute: ​

Machine Learning Properties (No 3D Required) ​

Property Prediction Speed Comparison ​

Force Fields & Molecular Mechanics ​

UFF (Universal Force Field) ​

MMFF94 (Merck Molecular Force Field) ​

Molecular Alignment ​

🌐 Multi-Platform Entry Points ​

Rust ​

Python ​

TypeScript / WASM ​

CLI ​

📖 Next Steps ​

What is sci-form?

Why sci-form?

🔗 The Complete Workflow

The Conformer Pipeline (ETKDGv2)

Pipeline Stages (9 Steps)

Automatic Retry Loop

Quantum Chemistry Methods

1️⃣ Extended Hückel Theory (EHT)

2️⃣ PM3 (NDDO Semi-Empirical SCF)

3️⃣ GFN-xTB (Generalized Tight-Binding, GFN0)

Comparison: EHT vs PM3 vs xTB

Quantum Chemistry (EHT)

Key Capabilities

From EHT, Compute:

Machine Learning Properties (No 3D Required)

Property Prediction Speed Comparison

Force Fields & Molecular Mechanics

UFF (Universal Force Field)

MMFF94 (Merck Molecular Force Field)

Molecular Alignment

🌐 Multi-Platform Entry Points

Rust

Python

TypeScript / WASM

CLI

📖 Next Steps