MoleBench user manual

1 · Overview

MoleBench is an in‑browser computational‑chemistry workbench. You build a molecule, see it in 3D, and run real quantum chemistry on it — nothing to install or configure. Light work (drawing, descriptors, pKa‑site detection) runs in your browser with RDKit; the heavy jobs run on our server with xtb (GFN2‑xTB), Psi4 (HF, DFT, MP2) and a GIAO engine for ¹H/¹³C NMR.

The Studio is a numbered flow. The left side is your structure (viewer + molecule library); the right side is a stack of analysis panels you work down in order: 1 Properties → 2 Quantum → 3 Spectra → 4 Orbitals → 5 Conformers & scans → 6 Reactions, plus the Export panel. Beyond the Studio there's a Protein workbench, a set of calculators, and 29 guided labs.

2 · Quick start

1. Type a name (aspirin) or a SMILES (CCO) in the build bar and press Build molecule — or click ✏️ Draw to sketch one.
2. Read the instant Properties (panel 1); click QSAR descriptors for the full descriptor set.
3. Press Optimize & compute (panel 2) for a real GFN2‑xTB geometry, energy, HOMO–LUMO gap and dipole — or DFT, Estimate pKa, or pick your own level under ⚙️ Advanced.
4. Open panels 3–6 to compute spectra (IR, UV‑Vis, NMR) + thermochemistry, render orbitals/ESP maps, search conformers or scan a bond, or drive a reaction to its transition state.
5. Use Export to copy the SMILES, share a link, download the 3D .sdf / 2D .png, or — if signed in — save the molecule to your account.

3 · Accounts & plans

MoleBench is free to start and needs no login to build molecules, read properties and use the calculators. Server‑side quantum jobs and the labs run on shared compute, so heavier use is organised into plans:

Plan	Price	What you get
Free	$0, no account	Build & view in 3D, all properties & descriptors, every calculator, plus a small daily allowance of server calculations.
Student	$10 / year	Everything free, plus unlimited server compute — DFT, NMR, orbitals, IR/UV‑Vis, conformers, reactions & transition states, pKa — and all 29 labs; save molecules to your account.
Pro	$120 / year	Everything in Student, with higher limits and priority for research‑scale work.

Sign in from the header chip (or you'll be prompted when you open a lab). Accounts are self‑hosted; payment is handled securely by Stripe and you can cancel anytime. Saved molecules live under 📁 My molecules in the Export panel.

4 · Building molecules

There are four ways to get a structure onto the bench:

• By name — common and IUPAC names resolve through PubChem (e.g. caffeine, 2-propanol).
• By SMILES — type any valid SMILES (e.g. c1ccccc1 for benzene). RDKit parses it and generates 3D coordinates.
• Draw it — the ✏️ Draw button opens a 2D sketcher (Kekule). Draw a structure and press “Use this structure”. You can also 📂 Import file (or drag‑and‑drop) a structure onto the canvas — a ChemDraw .cdxml file, or .mol / .sdf / .cml / SMILES. (ChemDraw’s XML .cdxml loads directly; its binary .cdx doesn’t — for that, use Save As → MDL Molfile or Copy As → SMILES.)
• Molecule library — the library card (left, below the viewer) has curated molecules grouped by category; click any to load it.

Whatever you build becomes the “current” molecule that every analysis panel acts on. Building a new molecule clears the previous results.

5 · The 3D / 2D viewer & measure

The viewer (top‑left) shows the structure. The 3D/2D tabs switch between an interactive 3D model (3Dmol.js) and a flat RDKit depiction.

Display styles

In 3D, the style chips switch between ball & stick, stick, sphere (space‑filling), line, and surface. The ◌ spin chip toggles auto‑rotation. Drag to rotate, scroll to zoom, right‑drag to pan.

📏 Measure

Click the 📏 measure chip, then click atoms in the model: 2 atoms → bond length (Å), 3 atoms → angle (°), 4 atoms → dihedral (°) — the standard distance / angle / dihedral measurement tools.

Highlight functional groups

Below the viewer, the Highlight group chips mark substructures (carbonyl, hydroxyl, amine, aromatic ring, etc.) on the 2D depiction using SMARTS matching — useful for teaching functional‑group recognition.

6 · Properties & descriptors (panel 1)

Computed instantly in your browser with RDKit, no server call:

Property	Meaning
Formula / MW	Molecular formula and molecular weight (g/mol).
logP	Calculated octanol–water partition (Crippen) — lipophilicity.
TPSA	Topological polar surface area (Ų) — passive‑permeability proxy.
H‑bond donors / acceptors	Counts used in drug‑likeness rules.
Rotatable bonds	Conformational flexibility.
Lipinski / Veber	Rule‑of‑five and Veber pass/fail flags (guidelines, not verdicts).

QSAR descriptors

The QSAR descriptors button expands the panel into a full medicinal‑chemistry descriptor set, grouped for readability — all instant, from the 2D structure:

• Size & weight — molecular weight, exact mass, heavy‑atom and heteroatom counts.
• Lipophilicity & polarity — cLogP (Crippen), molar refractivity, TPSA, Labute ASA.
• H‑bonding & flexibility — donors, acceptors, rotatable bonds, fraction of sp³ carbons.
• Rings & aromaticity — ring, aromatic‑ring, aliphatic‑ring and saturated‑ring counts.
• Topology & complexity — Bertz complexity, Balaban J, connectivity indices (χ), shape indices (κ).
• Drug‑likeness — the QED score (0–1), Lipinski rule‑of‑five violations, Veber pass/fail.

7 · Quantum calculation (panel 2)

The core compute panel. Several routes, all running real quantum chemistry on the server. A charge box (set ±1 for ions) and a solvent dropdown sit alongside the buttons.

Optimize & compute — GFN2‑xTB

Embeds a 3D structure, optimizes the geometry with GFN2‑xTB (a fast, accurate semi‑empirical tight‑binding method), and returns total energy (Hartree), HOMO–LUMO gap (eV), dipole moment (Debye) and atomic charges. The 3D view updates to the optimized geometry. The everyday workhorse — a few seconds per molecule.

DFT (B3LYP)

Runs a real B3LYP/6‑31G* density‑functional single point (Psi4) at the xTB‑optimized geometry, returning the DFT energy, HOMO/LUMO and dipole. Slower than xTB; small molecules.

Estimate pKa

The Estimate pKa button predicts the acidity of the most acidic site. For every ionizable O–H, N–H or S–H it optimizes the neutral acid and its conjugate base in water (xtb ALPB), takes the deprotonation energy, and maps it to a pK_a through an empirical calibration (fitted to 15 known acids, R² = 0.91). It reports the most acidic site plus a per‑site breakdown. Treat it as an estimate (±~1–2 units) — the trends are reliable.

Solvent (implicit solvation)

The solvent dropdown applies an implicit (continuum) solvent — xtb ALPB for xTB jobs and Psi4 PCM for DFT — to Optimize, DFT, IR and the reactions. Nine solvents are available (water, methanol, acetone, DMSO, acetonitrile, chloroform, CH₂Cl₂, THF, toluene). The reaction field stabilises charge: dipoles grow, acids strengthen, and anionic SN2 barriers rise versus the gas phase.

⚙️ Advanced — pick your own level of theory

The disclosure inside the panel lets you choose the method × basis × task yourself:

Choice	Options
Method	GFN2‑xTB · Hartree‑Fock · MP2 (correlated) · B3LYP · PBE0 · M06‑2X
Basis set	STO‑3G · 3‑21G · 6‑31G* · 6‑31G** · 6‑311G** · def2‑SVP · def2‑TZVP
Task	Single‑point energy · Geometry optimize

GFN2‑xTB runs on xtb; every other method runs on Psi4. MP2 adds electron correlation on top of Hartree‑Fock (the difference is the correlation energy). Higher levels and larger basis sets are more accurate but slower, so the advanced panel is capped to small molecules.

8 · Reactivity indices (conceptual DFT)

After a DFT run or an Orbitals run (anything that yields absolute HOMO and LUMO energies), MoleBench derives the conceptual‑DFT reactivity descriptors and shows them under panel 2:

Index	Definition	Tells you
Electronegativity χ	−(εHOMO+εLUMO)/2	Tendency to attract electrons.
Hardness η	(εLUMO−εHOMO)/2	Resistance to charge transfer; large = unreactive.
Chemical potential μ	(εHOMO+εLUMO)/2	Escaping tendency of electrons (= −χ).
Electrophilicity ω	μ²/(2η)	Global electrophilic power.

9 · Spectra & thermochemistry (panel 3)

IR + thermochemistry

Compute IR + thermo runs a GFN2‑xTB optimization plus a full Hessian (--ohess), then returns two things in one shot:

• IR spectrum — a stick spectrum of the vibrational modes (4000→0 cm⁻¹, IR convention), peaks scaled by intensity. Click any peak to animate that normal mode on the 3D model.
• Thermochemistry at 298 K — zero‑point energy, enthalpy H, Gibbs free energy G, entropy S, and heat capacity C_v, all from the same Hessian. These let you compute reaction enthalpies and free energies (ΔH, ΔG).

UV‑Vis (TD‑DFT)

UV‑Vis runs a time‑dependent DFT (TD‑B3LYP) excited‑states calculation and plots the absorption bands — excitation energies (nm), oscillator strengths and the dominant orbital transitions. Two tiers: a Quick run (STO‑3G) handles bigger molecules (up to ~26 atoms, approximate band positions) and an Advanced run (3‑21G) is more accurate but small molecules only (≤~14 atoms). Conjugation shrinks the HOMO–LUMO gap and red‑shifts the absorption — the orbital origin of colour.

NMR (¹H / ¹³C)

NMR predicts the ¹H and ¹³C spectrum with genuine GIAO B3LYP shielding tensors, then applies an empirical linear scaling against an experimental calibration. Two tiers: a Quick run (3‑21G, ~2–3× faster, up to ~30 atoms) and an Advanced run (6‑31G*, most accurate, ≤22 atoms) — each with its own calibration. Equivalent nuclei are grouped and integrated, and the result is shown as stick spectra with the peak list. Accuracy: ¹H within ~0.2 ppm, ¹³C within ~8 ppm across the full range (benzene ≈128.7 vs lit 128.4; a ketone carbonyl ≈206 vs 207). Very polar O–H/N–H protons still scatter (gas‑phase model, no hydrogen bonding).

10 · Orbitals, density & ESP maps (panel 4)

Choose a method (Hartree‑Fock or B3LYP) and basis (STO‑3G, 3‑21G or 6‑31G*), then press Compute orbitals (Psi4). An energy‑level diagram appears — click any level to view that orbital (HOMO−2 … LUMO+2) as a phased isosurface (the blue/red lobes are the orbital's ± phases). A row of chips switches what is painted on the model:

Chip	Shows
HOMO map / LUMO map	The frontier orbital as a coloured surface map.
Density	Total electron‑density isosurface (molecular “size/shape”).
ESP map	Electrostatic potential on the density surface — red = electron‑rich, blue = electron‑poor. The classic reactivity map.
Ionization map	Local ionization energy — red = electrons most loosely held (electron‑rich, electrophile‑directing). A short extra calculation over the occupied orbitals.
Localized (bonds)	Pipek‑Mezey localized orbitals — the rigorous version of Lewis bonds, lone pairs and cores, each labelled.
Spin density	For open‑shell radicals/ions (set charge ±1) — where the unpaired electron lives.
Off	Back to the plain model.

An isovalue slider tunes every surface. The panel reports the HOMO/LUMO energies and the gap, which feed the reactivity indices. Orbitals are heavier than xTB, so this is capped to small molecules (localized orbitals smaller still).

11 · Conformers & coordinate scans (panel 5)

Conformer distribution

Search conformers generates a set of 3D conformers (RDKit ETKDG), optimizes each with the MMFF94 force field, removes duplicates, and ranks them by energy with Boltzmann populations at 298 K. The lowest‑energy conformer is loaded into the 3D view. Use it to find the preferred shape of a flexible molecule (e.g. butane's anti vs gauche).

1D dihedral scan

Pick a rotatable bond and the dihedral scan drives it through 360°, re‑optimizing the rest of the molecule at each step (a relaxed scan), and plots the torsional energy profile. Click a point to load that geometry; download the curve as CSV. This is how you map a rotational barrier (e.g. ethane's ~3 kcal/mol threefold barrier).

2D conformational map (PES)

The 2D map drives two dihedrals together on a grid (MMFF94) and renders the result as a potential‑energy‑surface heatmap — basins (stable conformers), ridges (barriers) and saddle points in one picture, like a Ramachandran plot. Click any cell to load that conformer and read its relative energy. It reveals coupling a 1D scan misses, such as pentane's syn‑pentane clash.

12 · Reactions & transition states (panel 6)

This panel computes a reaction path and its transition state. A relaxed GFN2‑xTB scan drives the chosen bonds along the reaction coordinate; the energy maximum is the approximate TS and the path is an approximate IRC. Refine TS then converts that maximum into a rigorous saddle point in Psi4 (optking TS search + a frequency calculation at HF/3‑21G or HF/6‑31G*), confirmed by exactly one imaginary frequency whose normal mode is the reaction — scrub through it to watch the bonds break and form.

• Presets — five curated reactions: SN2, E2, Diels–Alder, the Menshutkin reaction, and an electrocyclic ring‑opening.
• ✎ Custom mode — on any molecule you've built, pick the bonds to form and/or break by clicking atoms in the 3D view, and the same scan + saddle‑refinement machinery runs your own reaction coordinate (best for intramolecular steps).
• The solvent picker applies here too — useful for reactions like Menshutkin that only proceed in solution.

The scan gives the right shapes and story fast (GFN2‑xTB); the Psi4 refinement makes them genuine first‑order saddle points — for learning, not benchmark barrier heights.

13 · Export, save & share

Copy SMILES	Canonical SMILES to the clipboard.
🔗 Share link	A URL that rebuilds the exact molecule (/tools/?q=…).
Download .sdf	The current 3D structure (optimized, if you ran a calculation) as an MDL molfile.
Download .png	The 2D depiction as an image.
💾 Save to my account	Stores the molecule against your account (Student/Pro). Requires signing in.
📁 My molecules	Re‑opens any molecule you've saved.

14 · Calculators

On the Studio page (scroll down, or the Calculators nav link) — the everyday chemistry math:

Equation balancer	Balances any reaction (e.g. C3H8 + O2 → CO2 + H2O) by solving the element matrix.
Reaction thermodynamics	From standard data (ΔHf°, S°, ΔGf° at 298 K for ~60 species) computes ΔH°rxn, ΔS°rxn, ΔGrxn (recomputed at any temperature via ΔG = ΔH − TΔS), the equilibrium constant K, spontaneity, and the crossover temperature where ΔG = 0. Include phases, e.g. CH4(g) + 2 O2(g) → CO2(g) + 2 H2O(l).
Molar mass & moles	Parses formulas with parentheses and hydrates (CuSO4·5H2O); converts grams ↔ moles.
Dilution C₁V₁=C₂V₂	Leave any one box blank — it’s solved.
Gas law PV=nRT	Leave any one box blank — it’s solved.
pH	pH/pOH of a strong acid or base from concentration.
Beer–Lambert A=εbc	Leave any one box blank — it’s solved.

15 · Protein workbench

The Proteins page is a structural‑biology workbench (3Dmol.js) for exploring proteins, enzymes, nucleic acids and their bound ligands. Load a structure three ways from the search bar:

• PDB ID — any Protein Data Bank entry (e.g. 1HSG, 4HHB, 1BNA).
• Keyword search — type a name like hemoglobin and pick from the top matches.
• AlphaFold (UniProt) — enter a UniProt accession (e.g. P69905) to load the AlphaFold predicted model.

What you can do

Panel	Function
Structure	Live metadata from RCSB — title, method, resolution, organism, and chain / residue / atom / ligand counts.
Representation	Show as: cartoon, surface, sticks, spheres, lines, ribbon/trace. Color by: rainbow, chain, secondary structure, element, B‑factor (= pLDDT confidence for AlphaFold models) or hydrophobicity.
Chains	List the chains and isolate any one (or Show all).
Ligands & binding sites	Each bound ligand has focus (zoom), pocket (residues within 5 Å as labelled sticks), contacts (polar H‑bonds with distances) and ⚛ Studio — which sends the ligand into the molecule Studio (its SMILES loads) so you can compute its properties, drug‑likeness or orbitals.
Display	Toggle water, hydrogens, residue labels, and a 📏 measure mode (click two atoms → distance).
Sequence	One‑letter sequence per chain; click a residue to highlight and zoom to it.
Saved views	Save the current scene (structure, representation, colour, isolated chain and exact camera angle) and restore it later.
Viewer toolbar	Spin, light/dark background, fit/reset, and export a PNG snapshot or download the PDB file.

16 · Labs

The Labs are 29 guided, multi‑step investigations that chain the tools together to teach a concept — each with learning objectives, background theory, a step‑by‑step procedure, a worksheet to fill in, questions and worked answers. Two tracks:

• Molecular chemistry (23) — polarity, drug‑likeness, conformers, the carbonyl IR stretch, bond order, VSEPR, alkene stability, ring strain, keto–enol, reaction thermodynamics, conjugation & colour, reactivity maps, localized orbitals, torsional barriers, substituent effects, acidity, carbocations, resonance, transition states, solvent effects, NMR assignment, 2D PES, and an acidity‑by‑design capstone (pKa + descriptors + MP2).
• Structural biology (6) — protein architecture (4 levels of structure), the hydrophobic core, enzyme active sites, structure‑based drug design, AlphaFold & the folding problem, and the DNA double helix — run in the protein workbench.

The labs are part of the Student & Pro plans (see §3).

17 · Methods & accuracy

Engine	Used for	Notes
RDKit	Parsing, 2D/3D embedding, descriptors, conformers, scans, SMARTS, pKa‑site detection	Runs in your browser (and on the server for conformers/scans).
GFN2‑xTB	Geometry optimization, energy, gap, dipole, charges, IR, thermo, pKa cycle, reaction scans, implicit solvent (ALPB)	Semi‑empirical tight‑binding — fast, surprisingly accurate; the default.
Psi4	HF, MP2, B3LYP/PBE0/M06‑2X, orbitals, density, ESP, UV‑Vis (TD‑DFT), TS refinement, PCM solvent	Full ab‑initio / DFT; small molecules only.
GIAO NMR	¹H/¹³C shielding (B3LYP — Quick 3‑21G / Advanced 6‑31G*)	Used for the ¹H/¹³C predictions, empirically scaled to experiment.

18 · Capabilities table

A summary of what MoleBench can compute and the method behind each result.

Capability	MoleBench
Build by sketch / name / SMILES	yes
3D viewer + display styles	yes
Measure distance / angle / dihedral	yes
MMFF geometry & conformer distribution	MMFF94, Boltzmann @ 298 K
Semi‑empirical	GFN2‑xTB
Hartree‑Fock	yes
MP2 (electron correlation)	yes (Psi4)
DFT	B3LYP, PBE0, M06‑2X
Basis sets to 6‑311G** / def2	yes
Geometry optimization & single point	yes
IR / vibrational frequencies	spectrum + click a peak to animate that normal mode
Implicit solvation	xtb ALPB (optimize/IR/reactions) & Psi4 PCM (DFT); 9 solvents
Thermochemistry (ZPE, H, S, G, Cᵥ)	yes
HOMO/LUMO, dipole, charges	yes
Molecular orbitals, density, ESP map	yes
Reactivity indices (χ, η, μ, ω)	yes
pKa estimation	GFN2/ALPB deprotonation cycle, empirically calibrated (±~1–2 units)
Drug‑likeness / QSAR descriptors	~25 descriptors incl. QED, complexity, connectivity & shape indices
Protein / structural‑biology workbench	PDB + AlphaFold; representations, colour schemes, chains, ligand binding sites, sequence, measure, ligand→Studio
NMR (¹H/¹³C)	real GIAO DFT (B3LYP — Quick 3-21G / Advanced 6-31G*), TMS-referenced, equivalent nuclei grouped + integrated
UV‑Vis (TD‑DFT)	yes
Transition states / reaction paths	scan + rigorous saddle (optking TS + Hessian at HF/3-21G or HF/6-31G*, 1 imaginary mode, animated) — 5 presets plus a custom mode where you pick the bonds to form/break on your own molecule
Coordinate scans (1D & 2D)	1D dihedral scan + 2D conformational map (PES heatmap, MMFF94, click-to-view)
Raman	on the roadmap

NMR uses genuine GIAO B3LYP shielding tensors (Quick 3-21G or Advanced 6-31G*), then applies an empirical linear scaling regressed against an experimental calibration set computed with the same pipeline. Advanced (6-31G*) accuracy: ¹H within ~0.2 ppm (R²=0.995), ¹³C within ~8 ppm (R²=0.984) across the full range — benzene comes out 128.7 vs the real 128.4, a ketone carbonyl ~206 vs 207. Very polar O–H/N–H protons still scatter (gas-phase model, no H-bonding).

19 · Limits & FAQ

• Size caps. To keep the shared server responsive, quantum jobs are size‑limited: xTB ≈120 atoms, IR ≈50, advanced/DFT ≈20–25, orbitals ≈30, conformers ≈60.
• Why is DFT slower than xTB? It solves the real electronic‑structure problem rather than a parameterized approximation. Start with xTB; reach for DFT when you need it.
• Do I need an account? Not to build molecules, read properties or use the calculators — those are free and need no login. Server‑side compute and the labs are part of the Student/Pro plans (see §3); free use includes a small daily allowance.
• Is my work stored? Calculations run on demand and aren't retained server‑side. Share links encode the molecule in the URL. If you're signed in, you can deliberately save molecules to your account.
• Are the pKa / NMR / spectra numbers exact? No — they're real calculations but approximate. pKa is an empirically‑calibrated estimate (±~1–2 units); NMR is GIAO + scaling; semi‑empirical geometries and spectra are good for trends and teaching, not publication‑grade values.
• It failed on my molecule. Very large, charged, or exotic‑element structures can exceed a cap or fall outside the method's parameters. Try a smaller analogue or GFN2‑xTB.

Start the guided tutorial →   Open the Studio