Where Elicit looks beyond research papers

The most important questions in drug development need much more than just the scientific literature.

Suppose you’re deciding whether TREM2 is a promising target for Alzheimer’s disease. You might want to know: How strong is the human genetic evidence? Is the biology consistent across experimental models? Who else is developing drugs against it? What do their patents actually cover? What molecules have shown meaningful activity, and in which assays?

The evidence exists, but it lives in different places. Papers tell you one part of the story; genomic databases, patents, chemistry databases, conference abstracts, and safety reports tell you others. Answering the question well means finding the right evidence, understanding what each source can and cannot establish, and synthesizing it into a coherent view.

Below are five of those sources, what each adds, and how you can access them.

1. Patents

Elicit can pull from major patent offices (including the USPTO, EPO, and WIPO), read the full texts of patents, and provide information such as assignee, inventors, and claims. This enables it to answer questions that the scientific literature may not: who owns a given composition or method, when they filed, and how broadly their claim reaches. Queries that can trigger a patent pull are those that touch on intellectual property or competitive position, such as "Who holds patents on inhibitors of [target] and what are the key claims covered?", or "What freedom-to-operate signals exist for [modality] against [target]?" It can also be triggered with the Patents skill.

2. Target–Disease Evidence

Elicit pulls from Open Targets, a consortium of EMBL-EBI, the Wellcome Sanger Institute, and five pharma companies that aggregates evidence linking human genes to diseases. This allows Elicit to surface structured genetic and clinical evidence that a standard literature search won't and show the evidence behind the association. Ask how strongly a gene is tied to a disease, and the agent can pull from it: "What are the best-supported genetic targets for endometriosis?" or "How strong is the human genetic support for TSKU in endometriosis?" It can also be triggered with the Open Targets skill.

3. Cancer Dependencies

For cancer dependencies specifically, Elicit can pull from DepMap, the Broad Institute's Cancer Dependency Map, which profiles cancer cell lines for gene expression and genome-wide CRISPR-knockout viability. This allows the agent to find out whether a gene is a selective vulnerability in one cancer type or something every cell needs. Figures are quoted directly from the DepMap release and cited, rather than paraphrased. Questions that can trigger this tool are oncology dependency questions, such as "Is [gene] a selective dependency in melanoma cell lines, or broadly essential?" or "Which cell lines should I validate [gene] in?" It can also be triggered with the DepMap skill.

4. Safety Reports

FAERS and MAUDE are the FDA's spontaneous report databases for drug adverse events and medical-device events, respectively, fed by reports from manufacturers, clinicians, and patients (Elicit will continue to support mining of AEMS as FAERS and MAUDE migrate there). These are often the only public records of what has been reported for a marketed product. Elicit resolves a product name and then returns report counts, breakdowns by reaction and outcome, and trends over time. Spontaneous reports have no denominator and no causal adjudication, so Elicit uses them to surface and size a signal, not to estimate how often an event occurs or whether the product causes it. Ask about a marketed product's safety record and the agent will pull from these databases: "What adverse events are most frequently reported for [drug], and is any signal growing?" or "What device malfunctions have been reported for [device]?" It can also be triggered with the openFDA skill.

5. Drug Chemistry

Elicit can pull from ChEMBL, EMBL-EBI's manually curated database of bioactive molecules, primarily extracted from the medicinal chemistry literature. Each record links a compound to a target through a specific assay and publication, with standardized potency measurements such as IC50, Ki, and Kd. Ask a medicinal chemistry question to trigger these pulls: "How potent is compound X against target Y?", or "What is the most potent inhibitor of target Y in ChEMBL?" It can also be triggered with the ChEMBL skill.