Building an Agent Harness for Life Sciences: Introducing Deep Life Sci

LangChain announced Deep Life Sci, an open source agentic assistant for clinical and lab scientists. The company said it built the tool on its Deep Agents harness as a template that organizations can adopt and modify for their own use cases.

According to LangChain, Deep Life Sci can access clinical trial records from more than 600,000 registered studies on ClinicalTrials.gov, 29 million paper abstracts through PubMed, and 12 million full-text articles on PubMed Central. It can review hundreds of documents at once by assigning them to sub-agents. Each agent comes with a LangSmith sandbox that allows it to run code for data analyses. Users can upload PDFs, images, tabular data files, bibliographic files such as RIS, and sequence files such as SMILES and FASTA for the agent to include in its work.

LangChain described example workflows. In one, a lab scientist uploads an RNA-seq or proteomics results table; the agent runs enrichment in the sandbox, searches the literature for prior evidence linking hits to the phenotype, separates well-described genes from novel ones, and returns a ranked table with supporting papers. In another, a clinical development or HEOR team can find published trials of the standard of care in an indication, extract endpoint value, N, population characteristics and follow-up duration, and produce a table and a forest-plot-style comparison. LangChain said users can also upload reference protocol documents and curate feedback and edits for clinical documentation.

LangChain stated that every Deep Life Sci run is logged end-to-end in the user's own LangSmith account, including literature searches, sandbox code, documents each sub-agent read, and the path to the answer, serving as an audit record. Deep Life Sci ships with a default evaluation set that can be modified as integrations are added and new use cases identified. LangChain said the code is open source, allowing organizations to add internal data integrations, use different frontier and open source models, and enforce guardrails and approval gates.