#33 · Single-cell
LLM agents grounded in raw single-cell data for annotation and perturbation prediction
The dominant thread is coupling LLM reasoning to explicit biological data and tools rather than parametric knowledge alone. scPilot grounds an LLM in raw scRNA-seq data and bioinformatics tools for step-by-step reasoning on annotation, trajectories and TF targeting. CellMaster applies an LLM-based agent to zero-shot scRNA-seq cell-type annotation, reporting 7.1% accuracy gains over CellTypist and scTab across 9 datasets. MechPert aggregates proposals from multiple LLM agents via consensus to predict unseen Perturb-seq responses, improving low-data Pearson scores. ChatDIA and the GEO metadata curation study apply zero-shot or open-weight LLM workflows to proteomics and public database metadata, outside single-cell analysis. What is new is less the agentic pattern itself than its anchoring to raw data and standardized analysis tools.
Preprints
scPilot: Large Language Model Reasoning Toward Automated Single-Cell Analysis and Discovery
scPilot grounds an LLM in raw scRNA-seq data and bioinformatics tools for step-by-step reasoning on annotation, trajectories and TF targeting, with the scBench…
CellMaster: Collaborative Cell Type Annotation in Single-Cell Analysis
CellMaster is an LLM-based AI agent for zero-shot scRNA-seq cell-type annotation, reporting 7.1% accuracy gains over CellTypist and scTab across 9 datasets.
MechPert: Mechanistic Consensus as an Inductive Bias for Unseen Perturbation Prediction
MechPert uses multiple LLM agents proposing directed regulators with consensus aggregation to predict unseen Perturb-seq responses, improving low-data Pearson…
ChatDIA: A zero-shot large language model workflow for targeted analysis of data-independent acquisition mass spectrometry data
ChatDIA is a zero-shot LLM workflow for targeted DIA proteomics peptide identification, matching DIA-NN accuracy on S. pyogenes and outperforming it on…
Supporting Metadata Curation from Public Life Science Databases Using Open-Weight Large Language Models
Open-weight LLMs applied to GEO metadata via API retrieval classify Arabidopsis ABA-treated RNA-seq projects (F1>0.98), outperforming keyword search for…
Topic history
| Date | Rank |
|---|---|
| #33 | |
| #40 | |
| #27 | |
| #27 | |
| #9 | |
| #2 | |
| #16 | |
| #6 | |
| #16 | |
| #2 | |
| #7 | |
| #25 | |
| #5 | |
| #1 | |
| #2 | |
| #1 | |
| #1 | |
| #15 | |
| #6 | |
| #30 |