Gene Autoannotator

Introduction / Purpose

An application that uses LLMs to summarize published literature on supplied genes into specified fields, in order to automate the curation of gene annotations.

High Level Overview:

1. Input: Gene identifier (locus or name).

2. Paper Retrieval: Query PubMed Central to obtain relevant research articles.

3. Section Extraction: Extract the abstract, results, and discussion sections from each paper.

4. Initial LLM Summarization: Submit each section to multiple large language models (mistral-nemo:12b, llama3:8b, gemma3:12b) to generate independent summaries.

5. Consensus Generation: Reconcile the outputs from the three LLMs using a tiebreaker model (phi4:14b) to produce a unified section summary.

6. Gene-Level Aggregation: Combine the consensus summaries across all papers using gemma3:12b to generate a final, aggregated gene annotation.

7. Output: Structured gene annotation for the input gene.

Installation / Requirements

Usage

Limitations / Considerations

References