---
title: Cell Entity Annotation (CEA) Overview
url: https://www.emergentmind.com/topics/cell-entity-annotation-cea
type: topic
---

# Cell Entity Annotation (CEA) Overview

Cell Entity Annotation (CEA) is not a single, uniform task across the literature. In Semantic Table Annotation, it seeks a mapping from table cells to ontology entities, formally \(f:E\to 2^{Ent}\) for the set of cell values \(E\) and a target knowledge graph such as DBpedia [2508.12868]. In single-cell transcriptomics, it can be viewed as assigning to each cell—or to each cell cluster—a discrete cell-type label drawn from a predefined ontology, or as producing a mapping \(\text{CEA}(x_i)\in\{1,\dots,K\}\cup\{\texttt{unknown}\}\) when open-set recognition is required [2304.02697; 2504.04698]. In microscopy, the same term has been used both for assigning a semantic class label to every pixel in a heterogeneous cell image and for assigning cell-type labels together with fine-grained morphological attributes to single-cell images [2312.00671; 2509.26185]. This terminological spread suggests that CEA is best understood as a family of entity-grounding problems whose common structure is the assignment of semantically meaningful labels to “cells,” but whose inputs, supervision, ontologies, and evaluation criteria differ substantially.

## 1. Semantic scope and core formalizations

In the table-annotation literature, CEA is defined over a table \(T\) with \(m\) rows and \(n\) columns, where \(E=\{e_{ij}\mid 1\le i\le m,\,1\le j\le n\}\) denotes the set of all cell values; the objective is to predict, for each cell \(e_{ij}\), a possibly singleton set of ontology entities \(E_{ij}=f(e_{ij})=\{e_1,e_2,\dots,e_b\}\) [2508.12868]. The same paper places CEA inside the broader Semantic Table Annotation task, alongside Column Type Annotation.

In single-cell transcriptomics, one formulation uses a normalized expression matrix \(X\in\mathbb{R}^{N\times G}\) for \(N\) cells and \(G\) genes, with Cell Entity Annotation defined as a mapping \(A:\{1,\dots,K\}\to L\) from clusters to cell-type labels, or equivalently as a probabilistic classifier \(f:\mathbb{R}^G\to\Delta^{|L|}\) with assignment \(A(x)=\arg\max_{\ell\in L} f(x)_\ell\) [2304.02697]. A more explicit open-set formulation defines \(\text{CEA}(x_i)\in\{1,\dots,K\}\cup\{\texttt{unknown}\}\), thereby making novelty detection part of the annotation problem [2504.04698]. Cross-domain annotation between scRNA-seq and snRNA-seq has also been formalized as learning from a labeled source domain \(D_s=\{(x_{s,i},y_{s,i})\}_{i=1}^M\) and an unlabeled target domain \(D_t=\{x_{t,j}\}_{j=1}^N\), where the unknown target label set \(\mathcal{Y}_t\subseteq\mathcal{Y}_s\) induces a partial-domain-adaptation problem [2511.08996].

In microscopy, CEA has two distinct formal meanings. One assigns a semantic class label to every pixel in a microscopy image containing a heterogeneous mixture of cell types, distinguishing it from instance segmentation by requiring both “where are the cells” and “what type is each cell” [2312.00671]. Another defines CEA as the automated assignment of both cell-type labels and fine-grained morphological attributes to single-cell microscopy images, yielding a multi-attribute output rather than a single class label [2509.26185].

## 2. CEA in semantic table annotation

For semantic tables, the central difficulties are semantic loss of cell values, strict ontology requirements, homonyms, spelling errors, and abbreviations [2508.12868]. The LLM-agent approach in “An LLM Agent-Based Complex Semantic Table Annotation Approach” addresses these with a ReAct-based agent that iteratively reasons with an LLM and invokes external tools when needed. Five tools are described: a Data Preprocessing Tool, a Column Topic Detection Tool, a Knowledge Graph-Based Enhancement Tool, a Context-Supported CEA Selection Tool, and a Context-Supported CTA Selection Tool [2508.12868].

The table-CEA workflow is explicitly context-sensitive. If the column header is non-semantic but cell values are informative, the system invokes Column Topic Detection first and then proceeds to CEA; if cells are empty or meaningless but the header is clear, it skips CEA entirely and performs only CTA with richer context; otherwise, for a fully semantic column, it goes directly to KG lookup, CEA selection, and CTA ranking [2508.12868]. The disambiguation step is phrased as selecting a candidate entity for a cell by using row context, the column name or inferred topic, and a ranked list of candidate entities returned by DBpedia lookup.

Efficiency is improved through distance-based reuse of existing annotations. The method defines a Levenshtein distance \(d_{lev}(s,t)\) recursively and reuses an existing annotation whenever
$$
d_{lev}(\text{cell},\text{annotated\_cell}) < 0.2\cdot \min(|\text{cell}|,|\text{annotated\_cell}|).
$$
Combined with deduplication, early stopping on high-confidence decisions, and caching of KG lookup calls, this reduces processing from 177,355 cells in a naïve pass to 60,341 distinct cells, yielding a \(\approx 70\%\) reduction in end-to-end annotation time and a \(\approx 60\%\) drop in total LLM tokens consumed [2508.12868].

Evaluation follows SemTab practice using
$$
P=\frac{|\text{Correct Annotations}|}{|\text{System Annotations}|},\quad
R=\frac{|\text{Correct Annotations}|}{|\text{Target Annotations}|},\quad
F1=\frac{2\cdot P\cdot R}{P+R}.
$$
On Tough Tables, the reported CEA performance is \(F1=0.843\) and \(Pr=0.845\); on BiodivTab, it is \(F1=0.900\) and \(Pr=0.930\) [2508.12868]. Because the method is built around dynamic tool selection rather than a fixed matching rule, a plausible implication is that table CEA increasingly treats entity linking as a context-conditioned reasoning task rather than a pure lookup problem.

## 3. CEA in single-cell transcriptomics: supervised, integrative, and cross-domain settings

In single-cell RNA-seq, a standard pipeline begins with normalization, feature selection, dimensionality reduction, and clustering, after which marker genes \(M_k\) are obtained by differential expression and annotation is expressed as
$$
A(k)=\arg\max_{\ell\in L} P(\ell\mid M_k,\text{meta}_k),
$$
where \(\text{meta}_k\) may include tissue, species, or developmental stage [2304.02697]. Prompt-based LLM annotation of clusters has been reported to use top markers and metadata, with benchmarking against existing tools such as scBERT, SingleR, and manual annotation [2304.02697].

Cross-domain CEA between scRNA-seq and snRNA-seq introduces a distinct challenge: the target label set may be only a subset of the source label set. ScNucAdapt is described as the first method designed for cross-annotation between scRNA-seq and snRNA-seq datasets, and it addresses both distributional differences and cell-composition differences with partial domain adaptation [2511.08996]. Its architecture has three modules: a shared encoder \(E\), a dynamic clustering and partial-match module, and a shared classifier \(G\). The encoder is a two-layer MLP,
$$
z=E(x)=W_2\,\mathrm{ReLU}(W_1x+b_1)+b_2,
$$
and the classifier is another two-layer MLP that maps \(z\) to \(\hat y\in\mathbb{R}^p\) followed by softmax [2511.08996].

The training objective combines source-domain cross-entropy and a cluster-wise alignment term based on Cauchy–Schwarz divergence:
$$
\mathcal{L}=\mathcal{L}_{cls}+\mathcal{L}_{cs}.
$$
The partial-match loss is defined over dynamically inferred target clusters \(X_{j,t}\) and matched source-label subsets \(X_{a_j,s}\),
$$
\mathcal{L}_{cs}=\sum_{j=1}^{\hat C} D_{CS}(X_{a_j,s},X_{j,t}),
$$
and dynamic split-and-merge rules, adopted from DeepDPM, determine the target-cluster count \(\hat C\) [2511.08996]. Only source classes with a matched target cluster contribute to \(\mathcal{L}_{cs}\), thereby ignoring nonshared labels and avoiding negative transfer. Reported accuracies include 90.24 on Bladder-immune (partial), 97.69 on Bladder-stromal (closed), 87.12 on Kidney unpaired (partial), 98.39 on Tumor-CLL (partial), 94.06 on Tumor-MBC (partial), and 99.78 on Mouse cortex (closed); removing either the CS-divergence loss or the dynamic clustering causes a substantial drop of 5–15 percentage points in accuracy across tasks [2511.08996].

A separate line of work addresses inconsistent label granularity across datasets. Motwani et al. formulate integrative cell type annotation with a finest-resolution set \(\mathcal{C}=\{1,\dots,L\}\) and dataset-specific binning functions \(f_k:\mathcal{C}\to\mathcal{C}_k\), where each observed label corresponds to a subset \(g_k(j)=f_k^{-1}(j)\) of fine types [2111.12149]. Instead of discarding coarse labels, the model directly uses
$$
P(Y_{(k)i}=j\mid x_{(k)i},z_{(k)i})
=
\frac{\sum_{\ell\in g_k(j)}\exp(\eta_{(k)i,\ell})}
{\sum_{v=1}^{L}\exp(\eta_{(k)i,v})},
$$
with a group-lasso penalty on gene coefficients and a ridge penalty on batch-specific parameters [2111.12149]. The resulting estimator is optimized by a blockwise proximal gradient descent algorithm. In simulation and real PBMC analyses, the integrative methods IBMR-int and IBMR-NG are reported to outperform subset and relabel baselines, often approaching an oracle estimator [2111.12149]. This suggests that, in transcriptomic CEA, label-resolution mismatch is not merely a nuisance variable but a modeling target in its own right.

## 4. Zero-shot, retrieval-augmented, and agentic CEA in single-cell analysis

Zero-shot single-cell CEA has been benchmarked as a prompt-based reasoning problem in which top marker genes are serialized into text and a model predicts the cell type directly or via zero-shot Chain-of-Thought prompting [2412.02915]. SOAR evaluates 8 instruction-tuned LLMs across 11 datasets. On the scRNA-seq “Nature” benchmark, GPT-4o reports an average BLEU of 51.79 in zero-shot mode, while Mixtral-8×22B rises from 16.65 to 28.19 under zero-shot CoT; in the multiomics “scACT” benchmark, cross-modality translation through a VAE alignment module maps ATAC into pseudo-RNA before top-\(k\) marker extraction [2412.02915].

Agentic variants explicitly couple LLM planning with tool use. scAgent defines \(\text{scAgent}=(P,A,M)\), where \(P\) is a planning module, \(A\) is an action space of scRNA models, modular Mixture-of-Experts LoRA adapters, embedding analysis, and incremental training procedures, and \(M\) stores reference datasets, vector embeddings in Milvus, and system history [2504.04698]. Its MoE-LoRA update is
$$
W=W_0+\sum_{i=1}^{n} g_i(x)\,B_iA_i,
$$
and novel-cell discovery uses Euclidean-distance outlier detection and cosine similarity thresholds. On the CG benchmark, scAgent achieves macro \(F1=0.8931\), reported as \(+6.73\) percentage points over the second-best scTab(10X), with 10× less data [2504.04698].

CellMaster frames zero-shot annotation as a collaborative loop involving a Hypothesis Generation Agent, a Marker Selection Agent, an Expression Analysis Module, and a Result Evaluation Agent, all operating on clustered AnnData inputs [2602.13346]. It defines a cluster-type confidence score \(S(c,\tau)\) by combining marker expression strength, specificity, and a dataset-wide \(z\)-score stabilization term. Across 9 datasets spanning 8 tissues, CellMaster reports an average CL score of \(0.602\pm0.058\) in automatic mode versus a best baseline average of 0.531, and \(0.717\pm0.050\) with human-in-the-loop refinement, with a subtype gain of \(+22.1\%\) and paired Wilcoxon signed-rank \(p<0.01\) after FDR correction [2602.13346].

Knowledge-graph retrieval is a second major direction. ReCellTy uses a graph-structured feature-marker database \(G=(V,E)\) with seven node types—Marker, FeatureFunction, CellName, CellType, TissueClass, CancerType, and GeneFamily—and five chained LLM-agent tasks: CellType Query, CellType Selection, Feature Query, Feature Selection, and CellType Annotation [2505.00017]. Reported gains include an average human-evaluation improvement of \(+0.18\) over four baseline models, a \(+0.21\) gain for DeepSeek-chat, and an overall semantic-evaluation improvement of \(+3.8\%\), with GPT-4o-mini reaching up to \(+6.1\%\) [2505.00017].

GATHER introduces a more formal retrieval setting for what it calls zero-shot Cell Entity Annotation, where the query is a hyper-entity gene set \(Q=S=\{g_1,g_2,\dots,g_n\}\) and the signal “emerges” from joint co-occurrence rather than any decisive single gene [2605.06403]. On a cell-centric biological knowledge graph with \(|V|\approx120K\), \(|E|\approx2.5M\), and approximately 14 relation types, GATHER performs global multi-source traversal, context-aware gene weighting, and topology-aware convergence scoring,
$$
\mathrm{Score}(t)=\sum_{h=1}^{k}\alpha_h\sum_{g\in S_h(t)} w_g.
$$
With \(k=2\) and \(K=10\), it reports exact-match accuracies of 27.45 on Immune and 59.64 on Lung, using only one LLM call per sample versus 2–61 calls for KG-RAG baselines, and improvements that are significant at \(p<0.01\) against the best KG-RAG baseline on both datasets [2605.06403].

Trustworthiness-oriented systems add explicit verification. CellTypeAgent uses an LLM to propose top candidate cell types and then verifies them with CellxGene-derived ranks based on scaled average expression, fraction of cells expressing each marker, and a tissue-agnostic global expression statistic [2505.08844]. The final score is
$$
\mathrm{score}(c)=r_c+R_e(c)+R_\rho(c)+R_{global}(c).
$$
The paper reports that CellTypeAgent consistently outperforms GPTCellType, CellxGene-only, and PanglaoDB, with representative average agreement scores of approximately 0.85, 0.78, 0.73, and 0.69, respectively [2505.08844]. A common concern in this line of work is hallucination; the verification step is explicitly designed to use LLM outputs as hypotheses rather than final answers.

## 5. Imaging-based CEA: semantic segmentation and multi-attribute labeling

In heterogeneous microscopy images, CellMixer defines CEA as per-pixel semantic labeling rather than instance detection [2312.00671]. The method avoids pixel-level annotation by training on homogeneous cell populations with image-level labels and synthesizing heterogeneous mixtures through mixup-style augmentation. Foreground masks are obtained from a homogeneous image \(I\) by computing Sobel gradients \(S_x,S_y\), the gradient magnitude
$$
G=\sqrt{S_x^2+S_y^2},
$$
followed by Gaussian smoothing, morphological erosion, adaptive thresholding, and assignment of the global class label to all foreground pixels [2312.00671].

The mixed training sample combines two crops \((I_i,M_i)\) and \((I_j,M_j)\) with coefficient \(\alpha\in[0,1]\), using \(\alpha=0.5\) in practice. CellMixer employs a DINOv2-pretrained Vision Transformer encoder and a Segmenter decoder head, with per-pixel logits over background and cell types and a Tversky loss to handle class imbalance [2312.00671]. On artificial mixtures, CellMixer improves over the baseline from 72.79/64.94 to 90.41/83.07 on Jurkat, from 88.20/80.44 to 96.68/89.87 on K562, and from 56.19/39.43 to 91.34/72.28 on PBMC, measured as mean pixel-accuracy and mean IoU; gains are also reported on real Jurkat+K562 and PBMC+Jurkat/K562 mixtures [2312.00671].

AttriGen broadens the imaging definition by treating CEA as simultaneous cell-type and morphological-attribute annotation [2509.26185]. Its dual-model architecture combines a VGG16 CNN for 8-way cell-type classification on the PBC dataset and a Vision Transformer for 11-way multi-attribute classification on the WBCAtt dataset, with fusion by concatenation of the 8-dimensional probability vector and the 11-dimensional attribute vector into a 19-dimensional profile [2509.26185]. The type branch uses multi-class cross-entropy,
$$
L_{type}=-\sum_{i=1}^{N}\sum_{c=1}^{8} y_{i,c}\log p_{i,c},
$$
while the attribute branch uses binary cross-entropy over the 11 attributes [2509.26185].

The PBC dataset contains 17,092 images across eight classes, and WBCAtt contains 10,298 images with eleven morphological attributes [2509.26185]. Reported performance includes 98.83% accuracy for VGG16 on PBC and a new state-of-the-art GAA of 94.62% for Swin-S on WBCAtt, with highest per-attribute accuracies of 99.81% for Granularity, 99.61% for Granule Type, and 99.29% for Granule Color [2509.26185]. The paper further reports annotation at \(\sim 20\) ms/image, 2.26 minutes total for 6,784 unlabelled PBC cells, a 1.48% drop relative to a human expert accuracy of 96.10%, and an estimated cost reduction of \(\sim 95\%\) [2509.26185]. Here, CEA is tied not only to classification but also to interpretability, via Grad-CAM and attention-map analyses.

## 6. Evaluation regimes, recurrent limitations, and research directions

Because CEA spans distinct tasks, the evaluation regime depends on the object being annotated. Table CEA commonly uses precision, recall, and \(F1\) [2508.12868]. Single-cell transcriptomic studies variously report exact-match accuracy, ancestor-match accuracy in the Cell Ontology DAG, overall accuracy, weighted \(F1\), macro \(F1\), agreement score, and ontology-based CL score [2605.06403; 2504.04698; 2505.08844; 2602.13346]. Imaging studies use mean pixel-accuracy, mean IoU, ordinary accuracy, and Global Average Accuracy [2312.00671; 2509.26185]. This suggests that cross-paper comparisons are only meaningful within a shared task definition and ontology regime.

The limitations reported in these literatures are also task-specific. For convergence-centric retrieval, missing or mis-grounded CellType nodes degrade performance, a fixed horizon \(k\) may miss longer but informative indirect paths, and retrieval parameters such as \(\alpha_h\) and \(K\) require tuning because there is no end-to-end training [2605.06403]. For LLM-based single-cell annotation, hallucination, prompt sensitivity, latency, and cost remain recurrent concerns; CellTypeAgent notes that literature-search augmentation often hurts performance and increases prompt length, while CellMaster and scAgent note reliance on large LLMs and careful prompt engineering [2505.08844; 2602.13346; 2504.04698]. In transcriptomic open-set settings, novel or ultra-rare cell types absent from verification databases cannot be validated directly [2505.08844]. In imaging, synthetic mixing assumes linear blending and does not model occlusion or depth ordering in tightly packed clusters, and very dense clusters remain challenging [2312.00671].

Several forward directions recur across the papers. GATHER proposes dynamic horizon and relation-type weighting, joint training of retrieval scoring parameters, integration with supervised classifiers such as scGPT, and application to other hyper-entity reasoning domains [2605.06403]. scAgent proposes extending the agent to multi-omic, spatial, and perturbation data via additional LoRA plugins [2504.04698]. SOAR recommends incorporating external cell-marker databases via RAG and extending multiomics evaluation beyond ATAC [2412.02915]. CellTypeAgent suggests dynamically weighting LLM and database evidence, expanding verification to multiple databases, and improving mixture handling through structured multi-label prompts and scoring [2505.08844]. Across these works, a consistent pattern is that CEA increasingly combines structured retrieval, ontology grounding, and adaptive reasoning, but the precise meaning of “entity” remains domain-dependent.

Source: https://www.emergentmind.com/topics/cell-entity-annotation-cea