---
title: 'scPilot: LLM-Enabled Single-Cell Analysis'
url: https://www.emergentmind.com/topics/scpilot
type: topic
---

# scPilot: LLM-Enabled Single-Cell Analysis

Searching arXiv for the scPilot paper and closely related single-cell reasoning context.
scPilot is a framework for automated single-cell RNA-seq analysis that uses large language models to perform what its authors call **omics-native reasoning**: the model converses in natural language while inspecting single-cell data summaries derived from raw expression matrices, invoking bioinformatics tools on demand, and producing a transparent reasoning trace that links biological claims to computational evidence [2602.11609]. The system targets three core tasks—cell-type annotation, developmental trajectory reconstruction, and transcription-factor targeting—and is paired with **scBench**, a benchmark suite of 9 curated datasets intended to evaluate this style of reasoning systematically [2602.11609].

## 1. Conceptual basis and problem formulation

The framework is motivated by the claim that single-cell analysis remains bottlenecked by human-only reasoning even when mature computational pipelines are available. In the formulation used by the paper, conventional workflows can be written as
\[
y = g_{\text{tool}}(X; \theta),
\]
where \(X \in \mathbb{R}^{G \times N}\) is the expression matrix, \(g_{\text{tool}}\) denotes a tool such as Scanpy or Monocle, and \(\theta\) denotes manually tuned hyperparameters [2602.11609]. The criticism is not that such tools are ineffective, but that the biological assumptions encoded in parameter choices and downstream interpretation are often implicit rather than auditable.

The paper contrasts this with code-generating tool agents, expressed as
\[
y = g_{\text{src}_K}\big(g_{\text{src}_{K-1}}(\cdots g_{\text{src}_1}(X)\cdots)\big),
\]
arguing that in those systems the causal chain between data evidence and biological conclusion is often fragmented across code, logs, and dialogue [2602.11609]. scPilot is presented as a response to this gap: not merely a tool runner, and not merely a summarizer, but a framework in which reasoning over omics evidence is the primary computational object.

The central formalism is **omics-native reasoning (ONR)**. The initial state is the raw data,
\[
S_0 = X.
\]
At reasoning step \(k\), the model emits a pair \((c_k, o_k)\), where \(c_k\) is a natural-language claim or justification and \(o_k\) is a primitive omics operator acting on the current state. State evolution is
\[
S_k = o_k(S_{k-1}),
\]
and the full trace is
\[
R = (c_1, o_1), \ldots, (c_K, o_K).
\]
The final answer is obtained from the terminal state \(S_K\) via \(y = h(S_K)\) [2602.11609]. The paper characterizes this as a “verbal + computational proof.” This suggests that scPilot’s main innovation is methodological: it reframes single-cell analysis as a sequence of inspectable claim–evidence transitions rather than as a one-shot prediction or opaque pipeline.

## 2. System architecture and workflow

scPilot comprises three principal components: a **Problem-to-Text Converter \(C\)**, a **Bio-Tool Library \(T\)**, and an **LLM Reasoner \(R_\phi\)** [2602.11609]. The converter exists because full expression matrices are too large for direct inclusion in an LLM context window. It constructs a task-specific semantic sketch
\[
\Phi_q : \mathbb{R}^{G \times N} \to S_q,
\]
where \(S_q\) is a compressed textual abstraction of the dataset tailored to query \(q\) [2602.11609]. The paper stresses that this conversion is algorithmic rather than learned.

Depending on the task, the semantic sketch may include cluster sizes, top marker genes, dotplot summaries, pseudotime or lineage graph structure, TF–gene candidate scores, and Gene Ontology overlap [2602.11609]. This is meant to preserve the information that a human analyst would inspect while discarding the full matrix.

The Bio-Tool Library wraps established single-cell software, including **Scanpy**, **Seurat via Reticulate**, **Monocle 3**, **pySCENIC**, and plotting utilities [2602.11609]. Each tool returns structured JSON outputs together with short textual descriptions. This design choice is technically important because it constrains the LLM to reason over machine-readable evidence rather than arbitrary unstructured logs.

The reasoner is instantiated with off-the-shelf models such as **o1** and **Gemini-2.5-Pro** [2602.11609]. The end-to-end workflow is summarized in the paper as
\[
X \xrightarrow{C} \text{Prompt} \xrightarrow{R_\phi} \{ \text{Thought}_k, \text{call}_k \}_{k=1}^K \xrightarrow{T} R_{1:K} \xrightarrow{R_\phi} y.
\]
Operationally, a natural-language query is converted into a reasoning loop in which the model proposes hypotheses, requests tool outputs, revises prior beliefs if necessary, and emits a final answer with an accompanying rationale [2602.11609].

The paper emphasizes three design principles across tasks: biological context is included explicitly in prompts; reasoning is iterative rather than one-shot; and gains are obtained without task-specific weight fine-tuning, relying instead on prompting structure and evidence integration [2602.11609].

## 3. Task design and reasoning procedures

scPilot supports three task families, each converted into an ONR workflow rather than treated as direct prediction.

For **cell-type annotation**, the inputs include cluster assignments, top-\(k\) marker genes per cluster with default \(k=10\), metadata such as tissue and species, and prior iteration outputs when applicable [2602.11609]. The textual representation used in scBench is “Scanpy-Leiden clusters + top-\(k\) marker genes per cluster (\(k=10\))” [2602.11609]. Annotation proceeds through a propose–filter–solve loop. The model first generates cell-type hypotheses from top differentially expressed genes and context, then proposes additional marker genes for unresolved clusters, then evaluates tool-generated dotplot-style evidence to determine which genes are informative or ambiguous, and which clusters should be stabilized or revisited [2602.11609]. The paper states that annotation uses a maximum of three reasoning iterations.

A canonical biological example is the PBMC case in which **NKG7** alone is treated as insufficiently specific, whereas the joint pattern **CD3D + GNLY + NKG7** is used to distinguish NK cells from cytotoxic T cells, especially when **CD8A** is weak; plasma-cell hypotheses are excluded when **SDC1** is absent [2602.11609]. This is used to illustrate that scPilot attempts multi-marker reasoning rather than single-marker lookup.

For **developmental trajectory reconstruction**, the inputs include clustered scRNA-seq data, top marker genes, developmental timepoint proportions per cluster, biological context, and PAGA- or Monocle-derived trajectory evidence [2602.11609]. The workflow first annotates clusters using top genes and timepoint distributions, then identifies the most likely root cluster, then constructs a directional tree over remaining clusters, and then performs a Monocle-guided self-audit in which missing progenitor relations, annotation mismatches, hierarchy errors, and implausible transitions are explicitly reconsidered [2602.11609]. The output is a nested trajectory tree together with refined cluster annotations and a rationale for root and branch assignments.

For **transcription-factor targeting / GRN prediction**, the inputs include a TF–gene pair, tissue context, pySCENIC or GRNdb candidate edges with motif support, and Gene Ontology overlap between TF and gene [2602.11609]. The model is prompted to recall TF biology, compare the candidate gene with known targets, and return both a reasoning string and a possibility score in \([0,1]\) [2602.11609]. The paper uses examples such as **Stat1 \(\to\) Irf7** and **Klf4 \(\to\) Muc5ac** as successful predictions, and **Usf2 \(\to\) Pigr** and **Fos \(\to\) Hmox1** as informative failures [2602.11609]. These cases are presented as evidence that ontology overlap and tissue context can both help and mislead the model.

This task structure suggests that scPilot’s contribution is not only the use of tools, but the explicit decomposition of biological inference into revisable subclaims.

## 4. Benchmark construction and evaluation methodology

The companion benchmark **scBench** contains **9 expertly curated datasets**: 3 for cell-type annotation, 3 for trajectory inference, and 3 for GRN prediction [2602.11609]. The benchmark is intended to measure what the paper terms omics-native reasoning rather than only generic LLM competence.

For **cell-type annotation**, the datasets are **PBMC3k** (2,638 cells \(\times\) 13,714 genes, 8 cell types), **Liver** (41,000 cells \(\times\) 2,000 HVGs, 31 cell types), and **Retina** (20,091 cells \(\times\) 19,719 genes, 9 cell types) [2602.11609]. Ground truth comes from author-provided labels. Evaluation is ontology-aware: predicted names are standardized to Cell Ontology IDs and scored as \(1.0\) for exact matches, \(0.5\) for partial matches via parent–child overlap, and \(0.0\) otherwise [2602.11609].

For **trajectory inference**, the datasets are **Pancreas** (36,351 cells, 17,327 genes, 4 timepoints, 14 trajectory nodes), **Liver** (44,010 cells, 2,000 HVGs, 4 timepoints, 15 nodes), and **Neocortex** (33,976 cells, 35,543 genes, 2 timepoints, 16 nodes) [2602.11609]. Ground truth is a manually curated lineage tree from the original study. The metrics are node Jaccard similarity,
\[
J(V_{\text{pred}}, V_{\text{gt}}) = \frac{|V_{\text{pred}} \cap V_{\text{gt}}|}{|V_{\text{pred}} \cup V_{\text{gt}}|},
\]
graph edit distance,
\[
\mathrm{GED}(G_{\text{pred}}, G_{\text{gt}}) = \min_{\gamma \in \Gamma(G_{\text{pred}}, G_{\text{gt}})} \sum_{e \in \gamma} c(e),
\]
and spectral distance,
\[
d_{\text{spectral}}(G_{\text{pred}}, G_{\text{gt}}) =
\left( \sum_{i=1}^{n} (\lambda_i^{\text{pred}} - \lambda_i^{\text{gt}})^2 \right)^{1/2}
\]
[2602.11609].

For **GRN prediction**, the tissues are **Stomach** (23 verified TF–gene edges, 46 total questions), **Liver** (71 verified edges, 142 questions), and **Kidney** (49 verified edges, 98 questions) [2602.11609]. Positives are validated against **TRRUST v2**, while negatives are sampled to avoid validated or SCENIC-supported edges [2602.11609]. The main metric is AUROC,
\[
\mathrm{AUROC} = \int \mathrm{TPR}(\tau)\, d\mathrm{FPR}(\tau),
\]
with
\[
\mathrm{TPR}(\tau)=\frac{\mathrm{TP}(\tau)}{\mathrm{TP}(\tau)+\mathrm{FN}(\tau)}, \qquad
\mathrm{FPR}(\tau)=\frac{\mathrm{FP}(\tau)}{\mathrm{FP}(\tau)+\mathrm{TN}(\tau)}.
\]
The paper also specifies the confusion-matrix form
\[
C =
\begin{bmatrix}
\mathrm{TN} & \mathrm{FP} \\
\mathrm{FN} & \mathrm{TP}
\end{bmatrix}
\]
[2602.11609].

The benchmark uses pre-specified termination conditions rather than allowing models to decide when to stop reasoning, which the authors frame as a reproducibility measure [2602.11609].

## 5. Empirical results

The paper reports that iterative omics-native reasoning improves average **cell-type annotation** accuracy by about **11%** over direct prompting [2602.11609]. In Table 2, **scPilot (o1)** achieves **0.518** on Liver, **0.792** on PBMC, and **0.728** on Retina, compared with **Direct (o1)** at **0.560**, **0.667**, and **0.474** respectively [2602.11609]. Table 3 reports that the best scPilot values include **0.792** for PBMC3k with **o1** and also **0.792** with **Gemini-2.0-Pro**, **0.763** for Retina with **Gemini-2.0-Pro**, and **0.518** for Liver with **o1** [2602.11609]. The largest median gain is reported on Retina at **+0.180**, with smaller gains on PBMC3k (**+0.042**) and Liver (**+0.024**) [2602.11609].

For **trajectory inference**, the paper states that **Gemini-2.5-Pro** under scPilot cuts trajectory graph-edit distance by **30%** versus one-shot prompting, and elsewhere summarizes an average reduction of **26%** [2602.11609]. Table 4 gives representative improvements. For **Gemini-2.5-Pro**, GED-nx improves from **8.33 to 5.00** on Pancreas, **8.00 to 3.33** on Liver, and **13.33 to 9.50** on Neocortex; Jaccard improves to **1.000** on all three; spectral distance improves from **0.453 to 0.310** on Pancreas and **0.388 to 0.199** on Liver, with a slight worsening on Neocortex from **0.977 to 1.052** [2602.11609]. The paper reports median improvements of **\(-2.0\)** in GED and **\(-0.14\)** in spectral distance across model–metric pairs [2602.11609].

For **GRN prediction**, the contribution summary reports an AUROC improvement of **0.03** over direct prompting, while the task-specific section reports an average **+0.098** across tissues and models [2602.11609]. Table 5 shows, for **o1**, Stomach improving from **0.827 \pm 0.002** to **0.873 \pm 0.004**, Liver from **0.753 \pm 0.001** to **0.760 \pm 0.000**, and Kidney from **0.777 \pm 0.000** to **0.797 \pm 0.001** [2602.11609]. The paper notes that **GPT-4o** shows the largest relative improvement, with average AUROC gain **+0.162** [2602.11609]. On the Stomach subset, Table 7 reports **scPilot (o1)** at **0.873 \pm 0.004**, exceeding graph-based baselines and direct **o1** at **0.827 \pm 0.002** [2602.11609].

Across the full benchmark, the paper states that scPilot wins in **87 of 108 total comparisons** [2602.11609]. This suggests broad but not universal improvement. The authors are explicit that scPilot does not dominate every dataset or model.

## 6. Interpretation, limitations, and future directions

A central claim of the framework is that it improves **auditability** by tying each conclusion to prompt stages, tool outputs, and explicit natural-language justifications [2602.11609]. The paper repeatedly contrasts this with pipelines that either produce only outputs or hide the rationale in code and intermediate files. In scPilot, the reasoning trace is treated as a first-class artifact. A plausible implication is that the framework is designed as much for expert inspection and debugging as for automation.

The paper also presents several ablations. Removing dataset-level metadata from PBMC3k reduces annotation accuracy from **0.792 to 0.688** for **o1**, **0.646 to 0.583** for **GPT-4o**, and **0.604 to 0.416** for **GPT-4o-mini** [2602.11609]. Shuffling GO annotations in GRN prediction lowers Stomach AUROC from **0.873 to 0.813** for **o1**, **0.800 to 0.710** for **GPT-4o**, and **0.697 to 0.617** for **GPT-4o-mini** [2602.11609]. Corrupting Monocle inputs worsens Liver trajectory metrics for **o1**, with Jaccard falling from **1.000 to 0.933**, GED-nx rising from **8.00 to 11.33**, and spectral distance rising from **0.567 to 0.593** [2602.11609]. These experiments are used to argue that the system is genuinely using biological context and tool outputs rather than ignoring them.

The framework’s limitations are described explicitly. Compression into semantic sketches may discard subtle rare-cell signals [2602.11609]. On complex datasets, deeper reasoning can lead to “overthinking”; the Liver annotation result with **o1**—**0.518** for scPilot versus **0.560** for direct prompting—is cited as a case where iterative reasoning amplified ambiguity among related developmental states [2602.11609]. Performance also depends on the quality of upstream tools, as shown by the Monocle and GO perturbation studies [2602.11609]. The paper notes that plausible but incorrect biological narratives remain possible, especially in GRN tasks when GO overlap is overinterpreted [2602.11609].

The open-source model results are presented as another practical limitation. **Gemma-3-27B** is reported as substantially slower and weaker than proprietary models; PBMC3k inference takes **135.7 s** on four A100 80GB GPUs, compared with **8.8 s** for a comparable GPT-4o task [2602.11609]. The authors conclude that fully on-premise ONR is not yet practical for most laboratories. They also report per-call runtimes for TF–gene prediction ranging from **3.75 s** for **GPT-4o-mini** to **30.61 s** for **Gemini-2.0-Pro**, with **o1** at **11.42 s** [2602.11609].

Future directions proposed in the paper include improved data compression for rare populations, larger-context ONR workflows, retrieval-augmented biological reasoning, stronger hallucination mitigation, and the incorporation of **wet-lab feedback** for experimental validation [2602.11609]. This suggests that scPilot is intended not only as an automation layer over existing pipelines, but as an initial template for a broader class of LLM-guided, evidence-grounded bioinformatics systems.

In the paper’s own positioning, scPilot is not merely an LLM wrapper for single-cell software. It is a framework that formalizes single-cell analysis as a sequence of inspectable biological claims and tool-grounded revisions, with scBench serving as the corresponding evaluation substrate [2602.11609].

Source: https://www.emergentmind.com/topics/scpilot