Overview
scPilot brings iterative, data-grounded reasoning to single-cell RNA-seq analysis. An LLM inspects data, invokes bioinformatics tools, explains its hypotheses, and revises them as new evidence becomes available. The framework covers cell-type annotation, developmental trajectories, and transcription-factor targeting.
The accompanying scBench evaluates these capabilities across nine curated datasets. The paper reports an 11% average annotation accuracy improvement with o1 and a 30% reduction in trajectory graph-edit distance with Gemini 2.5 Pro relative to one-shot prompting.