ELISA: Hybrid AI Agent Bridges scGPT and BioBERT for Single-Cell Discovery
A novel framework known as ELISA (Embedding-Linked Interactive Single-cell Agent) has been unveiled by researchers. This interpretable system merges scGPT expression embeddings with BioBERT-driven semantic retrieval and LLM-based interpretation, facilitating interactive exploration in single-cell genomics. By offering direct access to transcriptomic representations, it overcomes challenges in converting single-cell RNA sequencing (scRNA-seq) data into biological hypotheses while enhancing the interpretability of expression foundation models through natural language. An automatic query classifier directs inputs to various analytical processes, including gene marker scoring and semantic matching. The framework also includes modules for pathway activity scoring across 60 gene sets and ligand-receptor interaction predictions. This advancement, detailed in a paper on arXiv (identifier 2603.11872), marks a pivotal moment in utilizing AI for biological research, potentially expediting genomic hypothesis generation.
Key facts
- ELISA stands for Embedding-Linked Interactive Single-cell Agent
- It unifies scGPT expression embeddings with BioBERT-based semantic retrieval
- An automatic query classifier routes inputs to gene marker scoring, semantic matching, or reciprocal rank fusion
- Integrated modules perform pathway activity scoring across 60+ gene sets
- Ligand-receptor interaction prediction uses 280+ curated pairs
- The framework supports condition-aware comparative analysis and cell-type annotation
- The paper is available on arXiv with identifier 2603.11872
- The announcement type is 'replace-cross'
Entities
Institutions
- arXiv