Chemistry

What the workbench is

The Chemistry workbench gives you two tools that chemists normally pay a per-seat license for. The first is a structure editor in the style of ChemDraw, a full drawing surface for molecules that runs entirely in your browser. The second is a literature and patent search in the style of SciFinder, where you pick a compound or draw a fragment and see the papers and patents that mention it. Both are free, both keep your structures in your own folder, and neither needs an account or a server. The workbench lives at /chemistry and is one click away from the rest of the app.

The reason it works without a backend is that the chemistry itself runs in your browser. The drawing surface, the structure cleanup, the molecular formula and weight, the canonical SMILES and InChIKey, all of it is computed locally by chemistry libraries compiled to run in the page. Your structures are written to your data folder as standard MDL Molfiles, the same format ChemDraw and every other tool reads, so nothing is locked inside ResearchOS.

The molecule library

The left rail of the workbench is your molecule library. It lists every molecule in your folder, each row carrying a small rendered structure thumbnail, the name, and the molecular formula and average weight. This is the same signature left-rail layout the Sequences and Data Hub workbenches use, so clicking through your collection feels the same everywhere.

A collection selector at the top of the rail groups molecules by project, so you can narrow the list to one project, view the molecules not yet filed under any project, or open the library to everything at once. Each option carries a live count. Below the selector, a search box filters the list by name as you type, and a sort control orders the list by most recent or by name. Clicking a row opens that molecule in the detail view on the right. The rail can be collapsed to a thin edge when you want the detail view full width, and the divider between the rail and the detail view drags to set the rail width, which is remembered for next time.

Search by structure. Beyond the name filter, a search-by-structure mode finds molecules in your own library by chemistry rather than text. You type a SMILES or SMARTS query and pick one of two modes. Substructure match returns every molecule that contains that fragment, and Similarity ranks the library by Tanimoto score against your query, with the match shown as a percent on each result, so the analogs you already have surface even when their names give nothing away.

Right-click and bulk actions. Right-clicking a molecule in the rail opens quick actions to rename it, duplicate it, send it to a note, experiment, or method, or delete it. Selecting several rows with their checkboxes lets you delete a batch or file it all under a project at once.

Getting molecules into the library

There are three ways to add a molecule, surfaced as the New, PubChem, and Import actions at the top of the library rail.

Draw a new structure. The New action opens the structure editor on a blank canvas. You draw the molecule with bonds, rings, templates, and the periodic table, then save it to the library. The editor is the same surface described under Drawing and editing structures below.

Import from PubChem. The PubChem action opens a search box. Type a compound name and the search returns a grid of candidate compounds from PubChem, each with its structure, formula, and weight, so you can pick the right one rather than guessing from a name alone. Importing a candidate pulls its structure into your library and records its PubChem compound id alongside the locally computed identity, which is what lets the literature search find its linked papers and patents later.

Import a structure file. The Import action accepts MDL Molfiles (.mol), SDF files (.sdf, where each record lands as its own molecule), and SMILES files (.smi, .smiles). A plain .txt file is treated as one SMILES per line, so a column of structures comes in as a batch. ChemDraw's own .cdxml and .cdx files are recognized too, though they are not parsed directly, so the workbench asks you to export them as MOL or SMILES from ChemDraw first and import that. Everything is parsed in your browser, so the file never leaves your machine. A single-molecule import drops you straight onto the new molecule in the detail view.

The molecule detail view

Clicking a molecule in the rail opens its detail view. This is the fast browse surface, and it does not load the heavy editor, so moving between molecules is instant. At the top is a large rendered depiction of the structure. Below it is an identity table with the molecular formula, the average molecular weight, the canonical SMILES, and the InChIKey, all computed locally from the structure rather than stored blindly from whatever the source provided.

Quick copy actions put the canonical SMILES, the InChIKey, or a reference to the molecule on your clipboard, and a Send to action pushes the molecule straight into a note, experiment, or method. The molecule reference pastes into a note as a chip that links back to this molecule, so a synthesis note can point directly at the compound it describes. A linked-projects section shows which projects the molecule belongs to, with controls to add or remove a project link, a Referenced in panel lists everywhere the molecule is already cited across your notes, experiments, and methods, and a literature panel loads on demand. An Edit structure button opens the molecule in the full editor.

Alongside the core identity, the detail view shows a properties panel with calculated druglikeness numbers, the cLogP, the topological polar surface area, hydrogen-bond donors and acceptors, aromatic rings, and rotatable bonds, next to a Lipinski Rule-of-Five badge that flags at a glance whether the molecule sits inside the usual oral-druglikeness limits. For a molecule imported from PubChem, some of these descriptors are carried from PubChem itself, while the core identity is always recomputed locally.

Drawing and editing structures

The structure editor opens in a panel over the workbench, launched by the New action or the Edit structure button. It is a full molecular drawing surface with bond and ring tools, a template library of common scaffolds and functional groups, the periodic table for any element, charges and radicals, stereochemistry, and reaction arrows. You can paste a SMILES or a Molfile straight onto the canvas, and a structure cleanup pass tidies bond lengths and angles.

When you save, the structure is written to your folder as a Molfile and the identity facts are recomputed from what you drew, so the formula, weight, SMILES, and InChIKey always match the structure on the canvas. The editor is a heavier surface than the detail view because it loads a complete drawing engine, which is why browsing molecules uses the lightweight detail view and the editor opens only when you actually draw or edit.

The editor keeps a version history. Every time you save, it records a version, and a History tab in the editor rail lists them so you can look back at an earlier structure and restore it. This mirrors the version history notes get, so an edit that turned out wrong is never a dead end.

Literature and patents

The literature search is the free answer to the feature chemists pay SciFinder for. You pick a compound or draw a fragment, and it shows the papers and patents that mention it. It is assembled entirely in your browser from three public, no-key sources, so there is no server in the middle and nothing to log in to.

The per-molecule view lives in the detail panel. Opening the literature panel for a molecule pulls its linked PubMed papers and patent identifiers from PubChem, and its full-text chemical mentions from Europe PMC, the open European biomedical literature index. Each paper links out to its article page and each patent links to Google Patents. A common compound returns tens of thousands of results, so the panel ranks and paginates them and always shows the total rather than dumping everything. You can star a paper or a patent, and the star persists on the molecule, so the next time you open it your saved references surface as a one-click strip above the live results.

The Literature action in the library rail opens the same search as a standalone surface, so you can look up a compound by name without first adding it to your library. Below the name search sits a substructure patent search powered by SureChEMBL, which indexes compounds extracted from 28 million patents. You type a SMILES or SMARTS fragment, and it finds patent compounds that contain that substructure. The search runs asynchronously on SureChEMBL, submitting the fragment and polling until the results are ready.

How identity is computed

Every molecule carries a set of identity facts, the molecular formula, the average molecular weight, the canonical SMILES, and the InChIKey. These are not taken on faith from the source of the structure. They are computed locally from the Molfile by RDKit running in your browser, so a structure you drew, one you imported from PubChem, and one you loaded from a file all get their identity the same way. The InChIKey in particular is a stable hash of the structure, which is what makes it a reliable key for looking a compound up in an external database.

Connection to the rest of the app

Molecules participate in the shared folder structure alongside notes, experiments, and sequences. A molecule is filed under a project through the linked-projects control in its detail view, and each project surface carries a Molecules section listing the structures linked to that project, so the compound shows up in the same project context as the experiments that use it.

Because a molecule has its own reference, you can mention it from a note. The molecule reference pastes as a chip that deep-links back to the structure, opening it in the workbench when clicked, the same way a note can reference a sequence or an experiment. This keeps the structure and the writing about it in one connected place rather than as a screenshot pasted into a document.

A molecule also drops into the figure composer at /figures as a panel, depicted by the same renderer the workbench uses, so a structure sits in a publication layout next to your plots and sequence panels. Chemistry panels carry no per-panel styling controls, by design, since the depiction is the structure itself.

Working with BeakerBot

BeakerBot, the assistant that runs throughout the app, can operate the Chemistry workbench for you. It can create a molecule from a SMILES you give it, pull one in from PubChem by name or CID, read a molecule's identity back to you, rename it, edit its structure from a SMILES you provide, and delete it to the Trash. When you have a molecule selected in the rail, it resolves "this molecule" to that selection, so you can ask it to rename or read the one you are looking at without naming it.

The line BeakerBot will not cross is inventing chemistry. It operates the app and relays what the on-device engine computes, so a formula, weight, or canonical form always comes from RDKit, never from the model's memory, and a structure edit only ever uses a SMILES you actually provided. Every write shows a preview first, and a delete asks for a confirmation, so nothing changes in your library without your say-so.

What it is built on

The Chemistry workbench stands on open-source chemistry. The drawing surface is Ketcher from EPAM, the structure perception and identity calculations are RDKit, and the literature search draws on PubChem, Europe PMC, and SureChEMBL. All of it runs in your browser against public data, which is what keeps the feature free and your structures local. These projects and the rest of the open-source software ResearchOS is built on are credited on the open-source page.