How to use the PubChem database for safety and toxicity data

The PubChem database is the US National Library of Medicine's open chemistry database, run by NCBI at NIH and free for anyone to use. It holds more than 124 million compounds and about 349 million substance records, drawn from over 1,000 data sources such as government agencies, chemical suppliers, pharma companies and journal publishers.

For regulatory and EHS work, start with Safety and Hazards, GHS classification and the older HSDB toxicology records. Ethanol is the worked example below.

Compound, substance and BioAssay records

PubChem started in 2004. It is split into three linked databases:

One compound (CID) can carry many substance records (SIDs) from different depositors. Its compound page brings together information from those sources. NLM's introduction explains how the three fit together.

Searching the PubChem database

You can search by name, molecular formula, structure or another identifier, and also by gene, protein, pathway, organism, cell line or patent number. Typing "ethanol" takes you to ethanol's Compound Summary (CID 702). The same works for methanol, caffeine or water.

A Compound Summary page is long. Ethanol has 20 top-level sections, including Names and Identifiers, Chemical and Physical Properties, Safety and Hazards, Toxicity, Literature and Patents. Use the Contents menu at the top to jump to the one you need. Pages only show sections that have data, so a rare compound will have far fewer.

Structure search and the PubChem Sketcher

If you don't know the name, draw the molecule. Open the PubChem Sketcher with "Draw Structure" on the homepage, which since the December 2025 interface update is a hexagon icon inside the search box. You can also go straight to the Sketcher page.

Chemical structure searching offers four main modes.

• identity, to find the exact structure
• similarity, which uses a 90% threshold by default
• substructure, to find compounds that contain your drawing
• superstructure, to find compounds that your drawing contains

Quick Search also offers 3D similarity.

Where the safety and hazard data sits

Ethanol's Safety and Hazards section includes first aid, fire fighting, handling and storage, exposure controls and transport information. These resemble the headings on safety data sheets (SDS and MSDS). But PubChem is not an SDS library. Use it for background, then read the supplier SDS for the product you actually handle.

GHS classification

The GHS Classification sits under Safety and Hazards, then Hazards Identification. For benzene (CID 241), the sources cited include ECHA, HSDB, NIOSH, OSHA, ILO-WHO ICSC, NITE-CMC and CAMEO.

Benzene's entry is aggregated "per 1934 reports by companies from 47 notifications to the ECHA C&L Inventory". The percentage beside a code is the share of companies that provided hazard codes and reported that code. Only codes above 10% are shown. PubChem warns that notifications vary with impurities, additives and other factors.

Some substances show "Not Classified". L-isoleucine was reported as not meeting GHS hazard criteria by 169 of 171 companies.

PubChem's GHS reference page follows GHS Rev.11 (2025). It lists the nine pictograms (GHS01 to GHS09), H and P statements, and older codes that are kept and marked. You can download the table as TSV.

LCSS

The Laboratory Chemical Safety Summary is a short view of health and safety data for one compound. It covers flammability, toxicity, exposure limits and symptoms, first aid, handling and clean-up. The layout follows the NRC's "Prudent Practices in the Laboratory" (2011), and it only appears for compounds that already have a GHS classification: more than 167,000 records as of 2023.

HSDB toxicity data

The Hazardous Substances Data Bank used to live on TOXNET, which NLM retired on 16 December 2019. HSDB content moved to PubChem and is now spread across the normal page sections. To see only HSDB entries, filter the page by source, or start from the HSDB source page.

An HSDB record, such as the one for acetophenone, has headings like Toxicity Summary, Evidence for Carcinogenicity, Human Toxicity Excerpts, Non-Human Toxicity Values, Ecotoxicity Values and Occupational Exposure Standards. Each entry cites its original reference. Go back to that reference before you quote a value in a dossier.

Downloading PubChem data

There are three ways to download: a single summary page, a script, or a bulk file. On a Compound Summary page, the Download button above the Contents menu gives the page data, the 2D structure and a 3D conformer. That conformer is computed, not an experimental structure. Each section also has its own Download button.

From a results list, the Download button on the right exports a summary, structure records or images. Summaries come as CSV, JSON, JSONL or XML, with fields you pick.

For scripts, use PUG REST. This URL returns aspirin as CID 2244, formula C9H8O4: https://pubchem.ncbi.nlm.nih.gov/rest/pug/compound/name/aspirin/property/MolecularFormula/JSON. Keep to 5 requests per second per user. Under heavy load the limit tightens and extra calls get HTTP 503. For the full annotation text, including safety sections, PUG-View uses /rest/pug_view/data/compound/{CID}/JSON. Large pulls belong on the bulk files (Compound, Substance, Bioassay, Patents), and NCBI prefers HTTPS for those downloads.

How far to trust it for compliance work

PubChem compiles what depositors report. It isn't a legal classification register. For a binding EU classification, check the harmonised entries in CLP Annex VI, not PubChem's aggregated notifications. NCBI does not warrant the accuracy, completeness or usefulness of the information, and it places no restrictions on use, though submitters may claim rights in their data. Each source has its own licence note. ILO ICSC is CC BY 4.0, while ECHA points to its legal notice. Check that note, and follow the cited reference, before you copy a value into a dossier. For product-specific safety information, read the supplier SDS.

RegAffairs AI's answer about the PubChem database cites NLM guidance and HSDB records.