AI for chemistry research should tell you where a number came from. In AI for Research, Deep Research has a Chemistry source group that looks up PubChem compound records and ChEMBL bioactivity data, and writes a cited answer that puts those records next to journal papers and preprints. A reported IC50 value points back to the ChEMBL record, and the record points to the assay and the paper.
Sources for chemistry and materials
| Source | What it holds | Where in the product | Credits |
|---|---|---|---|
| PubChem | Compound records: identifiers, names and synonyms, computed properties | Deep Research, Chemistry | As the run's depth |
| ChEMBL | Bioactivity measurements against targets, curated from the literature | Deep Research, Chemistry | As the run's depth |
| OpenAlex and Semantic Scholar | 250M+ scholarly works, including chemistry and materials journals | All papers source; Find Papers | Search 0 |
| arXiv and other preprint servers | Preprints, flagged as not peer reviewed (ChemRxiv is not listed among the sources) | Preprints source | As the run's depth |
| PubMed and Europe PMC | Medicinal chemistry and pharmacology articles | Medical source | As the run's depth |
| My Library | Your own PDFs, including papers behind your institution's access | My Library source | As the run's depth |
Depth costs are Summary 1 credit, Deep 15 and Detailed 45 (paid plans). CAS SciFinder and Reaxys are not searched.
Tools for chemistry and materials
The Chemistry group works best for questions that name a compound or a target: what has been reported, in which assays, and in which papers. For broad questions about a reaction class or a materials system, the paper sources do most of the work, and the answer says so by citing papers rather than records.
| Tool | What it does | Credits |
|---|---|---|
| Deep Research | Cited answers from PubChem, ChEMBL, papers and preprints | Summary 1, Deep 15, Detailed 45 (paid) |
| Find Papers | Search the literature with filters for year and journal tier | 0 |
| Journal formatter | Section order, word and abstract limits and reference style for a journal, plus what still breaks its guidelines | 3 per 10,000 words, at most 12 |
| Citation styles | ACS among 10,000+ styles in the editor | 0 |
| Data Lab | Regression, ANOVA and t-tests on your measurements, calculated in the browser | 0; AI write-up 2 |
| Scientific diagrams | Process flowcharts and study designs in the editor | 0; Draw with AI 1 |
| Reporting guideline check | SAMPL for statistical reporting | 5 per 10,000 words |
A worked example
Example (invented for illustration): a PhD student in medicinal chemistry is writing the introduction to a paper on a series of kinase inhibitors.
- Deep Research, sources Chemistry and All papers, Summary depth: "What bioactivity has been reported for the parent compound against its main kinase target, and in which assays?" The answer cites the PubChem record and four ChEMBL activity records with their assay types, and three papers. 1 credit.
- A follow-up asks which close analogs have published activity on the same target. 1 credit.
- One Deep report reviews selectivity problems for this inhibitor class: 15 credits.
- Her dose-response summary values go into Data Lab for a one-way ANOVA across the series, free.
- She writes the paper in the editor with ACS references, then runs the journal formatter on the 6,000-word draft: 3 credits.
Total: 20 credits, inside the Free plan's 45 a month.
Citation styles and reporting standards
Most chemistry journals use ACS style, and many materials journals use numbered styles from Elsevier, Wiley or RSC; the editor formats references in 10,000+ citation styles, ACS and RSC among them. When you cite a compound or an activity value, give the PubChem CID or the ChEMBL identifier as well as the paper, so a reader can trace it. For statistical reporting, the SAMPL guidelines in the guideline check apply to dose-response and assay comparisons as much as to clinical work.
What it does not do
- No molecular modeling: no docking, molecular dynamics or quantum chemistry.
- No virtual screening, QSAR models or property prediction.
- No structure or substructure search; it looks up records by name and identifier.
- No materials property databases. Crystal structures, band gaps and similar materials data are not looked up; materials papers come through the normal paper sources.
- No reaction planning or retrosynthesis.
- It does not check safety data for laboratory use. Read the safety data sheet and follow your institution's rules.
- Bioactivity values vary with the assay. Compare like with like, and read the original paper before you cite a number.
Who uses it
- PhD students and postdocs in chemistry: see AI for PhD students.
- Academic groups and supervisors: see AI for academics.
- Pharma, chemicals and materials R&D teams: see AI for industry.
For genes, proteins and pathways, see life sciences. For engineering and computing sources, see engineering and computing.
Learn more
- How to find sources for a research paper
- How to format a paper for a journal
- How to write a research paper
- What is deep research AI?
Last updated 2026-11-13