AI-powered Immune Cell Knowledge Graph (ICKG) with granular immune contexts enables immune program interpretation
Open accessAbstract
The widespread application of single-cell and spatial omics to models and patient samples has transformed immune cell profiling across physiological conditions. However, knowledge of immune cell states, functions, and gene regulation remains fragmented across publications, limiting our ability to synthesize insights and derive mechanistic understanding from the literature. To address this gap and facilitate literature integration, we constructed Immune Cell Knowledge Graphs (ICKGs)—four cell type-specific graphs derived from over 24,000 cancer immunotherapy-focused PubMed abstracts using large language models (LLMs) with “human verifiable” validation. Unlike conventional databases, which provide context-agnostic pathways, ICKGs capture directed, literature-supported relationships among genes, pathways and immune functions, enabling context-aware reasoning. We validated ICKGs using perturbation datasets from cytokine stimulation and CRISPR experiments, demonstrating that ICKGs contain more accurate and immunologically coherent contexts than canonical databases. As a key application, ICKGs provide interpretable and accurate pathway annotations, including signatures unannotated by canonical databases or used in immuno-oncology. To support community use, we created an interactive portal ( https://kchen-lab.github.io/immune-knowledgegraph.github.io/ ) to perform ICKG-based pathway annotations, allowing researchers to explore immune cell-specific insights grounded in literature. This work establishes ICKGs as a scalable framework for immune-specific functional interpretation and mechanistic hypothesis generation in single-cell and spatial omics.
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
LLMs can identify relevant literature given mechanistic context, read scientific text, and extract information into structured templates.
“LLMs can identify relevant literature given mechanistic context, read scientific text, and extract information into structured templates [14, 15, 16, 17, 18].”
From Structured Schemas for Provenance‐Rich, <scp>LLM</scp> ‐Assisted <scp>QSP</scp> Model Calibration (2026) · cites this paper as [14]
The Methods section on relationship extraction describes an LLM reading PubMed abstracts and classifying entity relationships into categories used to build a structured knowledge graph. The paper does not show the LLM identifying relevant publications from a mechanistic context; it works from a collection of abstracts. Its Discussion also specifies that the analysis used abstracts, not full texts. (LLM, conf=0.84)
Experimental — accuracy not yet measuredUnlike knowledge graph approaches that prioritize coverage across thousands of abstracts and can tolerate some extraction errors, parameter calibration requires higher per-target stringency because each value directly influences model predictions.
“Unlike knowledge graph approaches that prioritize coverage across thousands of abstracts [14, 32] and can tolerate some extraction errors, parameter calibration requires higher per‐target stringency because each value directly influences model predictions.”
From Structured Schemas for Provenance‐Rich, <scp>LLM</scp> ‐Assisted <scp>QSP</scp> Model Calibration (2026) · cites this paper as [14]
Split into 2 atoms — 0 confirmed, 2 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.
Experimental — accuracy not yet measuredLLMs can identify relevant literature given mechanistic context, read scientific text, and extract information into structured templates.
“LLMs can identify relevant literature given mechanistic context, read scientific text, and extract information into structured templates14,15,16,17,18.”
From Structured Schemas for Provenance-Rich, LLM-Assisted QSP Model Calibration (2026) · cites this paper as [14]
Unlike knowledge graph approaches that prioritize coverage across thousands of abstracts14,32 and can tolerate some extraction errors, parameter calibration requires higher per-target stringency because each value directly influences model predictions.
“Unlike knowledge graph approaches that prioritize coverage across thousands of abstracts14,32 and can tolerate some extraction errors, parameter calibration requires higher per-target stringency”
From Structured Schemas for Provenance-Rich, LLM-Assisted QSP Model Calibration (2026) · cites this paper as [14]
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
The citation lists haven’t been fetched yet
References 37 · Cited by 3
Research works printed but not linked: 2
- [3]Nature Publishing Group. Tinkering with immune cells gives cancer treatment a boost. Nature News. https://www.nature.com/articles/d41586-023-03747-3 (2023).
- [21]ACL 2013 BioNLP Shared Task. in: 2013 Proc. of the Workshop (Association for Computational Linguistics, Sofia, Bulgaria, 2013).
- Authors
- He, S., Tan, Y., Ye, Q., M Gubin, M., Dede, M., Rafei, H., Peng, W., Rezvani, K., Mohanty, V., & Chen, K.
- Affiliations
- The University of Texas MD Anderson Cancer Center (US)
- University of Houston (US)
- Corresponding author
- Mohanty, Vakul
- Chen, Ken
- Venue
- npj Artificial Intelligence
- Published
- 2026-01-27
- Volume
- 2
- Issue
- 1
- Pages
- 13-13
- ISSN
- 3005-1460 (ISSN-L)
- PMC
- PMC12846910
- OpenAlex
- W7125798115
- Cited by
- 3
- References
- 37
- Citations by year
- 2026: 3
- FWCI
- 6.20
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Locations
- npj Artificial Intelligence (publishedVersion, cc-by)
- PubMed (publishedVersion)
- PubMed Central (publishedVersion, cc-by)
- Digital Commons-TMC (Texas Medical Center) (submittedVersion)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
- Funders
- Cancer Prevention and Research Institute of Texas (RR190017)
- National Cancer Institute (R01CA282027, 5U01CA281902)
- Chan Zuckerberg Initiative (2024-345892)
- Topics
- Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Biomedical Text Mining and Ontologies
- Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Bioinformatics and Genomic Networks
- Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Single-cell and spatial transcriptomics
- Keywords
- large language models, cancer immunotherapy, pathway annotation, immune profiling, literature integration, gene expression regulation, immune cell states, single-cell omics, spatial omics, knowledge graph validation, CRISPR perturbation, cytokine induction, hypothesis generation
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
