LinkedOmicsChat is a natural language interface to cancer multi-omics databases. Ask research questions in plain English and get structured, cited answers from real data sources.
Data Sources
Results are grounded in live API calls to these datasets and services. Different workflows query different sources in real time depending on the question.
Core multi-omics and analysis services used by the app, including TCGA survival, CPTAC-integrated expression and correlation analyses, drug target annotations, and treatment-response studies served through LinkedOmics-hosted APIs.
Functional proteogenomic interaction network for discovering gene co-functional neighborhoods and pathway modules.
Gene set enrichment analysis across GO, KEGG, WikiPathways, and other ontology databases.
Underlying proteomics and phosphoproteomics datasets used in LinkedOmics-supported proteogenomic analyses across 10 cohorts.
NCBI literature search integrated into the research workflow for evidence-backed answers.
Identifier normalization service used to resolve Ensembl and UniProt identifiers to HGNC gene symbols before analysis.
Example Queries
These are representative questions across analysis types. You can also click example chips on the welcome page to launch them directly.
Architecture
LinkedOmicsChat uses a LangGraph agent loop with MCP-connected tool servers. The LLM autonomously decides which tools to call and in what order — enabling multi-step analyses without manual workflow configuration.
You type a question
Natural language query with no required syntax — gene name, cancer type, and analysis type are inferred from context.
LangGraph plans the steps
A LangGraph agent decides which tools to call, in what order, and with what parameters — including multi-step chains.
MCP tools query real APIs
Each tool calls the relevant live endpoint or service layer, including LinkedOmics-hosted APIs, FunMap, WebGestalt, PubMed, and MyGene.info.
LLM synthesizes the answer
Results are formatted into a cited markdown response with visualizations, tables, and literature references.
Tech stack
FAQ
Do I need an account to use LinkedOmicsChat?
No. You can try the platform as a guest — click any chip on the welcome page or use the 'Try as guest' button. Create an account to save session history and export results.
What cancer types are supported?
TCGA survival analysis supports 35 cohort codes, including aggregate cohorts such as COADREAD, GBMLGG, KIPAN, and STES. CPTAC-integrated expression, survival, and correlation analyses currently cover 10 cohorts: BRCA, COAD, CCRCC, GBM, HNSCC, LSCC, LUAD, OV, PDAC, and UCEC.
Can it run multi-step analyses automatically?
Yes. The LangGraph agent can chain tools without you specifying the steps. For example: 'Find BRCA1 neighbors and run enrichment' automatically calls FunMap, collects the gene list, then calls WebGestalt with those genes.
What if my question is ambiguous?
The assistant will ask a focused clarification question rather than guessing. For finite choices (like analysis type), it shows quick-select option chips. For open-ended inputs like gene names, it asks directly.
Is the data real or simulated?
Production responses use live API calls to LinkedOmics-hosted services, FunMap, WebGestalt, PubMed, and MyGene.info as needed.
Ready to start exploring your research questions?
Built by Zhang Lab · Data and services from LinkedOmics, TCGA, CPTAC, PubMed, and related research resources