TL;DR
FAVOR 2.0 is a major upgrade to a comprehensive online resource for annotating and interpreting genomic variations. It now offers vastly expanded annotations, a robust API, an interactive genome browser, and a novel AI-powered chatbot (FAVOR-GPT) to make complex genomic data more accessible and actionable for researchers.
What Problem Does This Paper Address?
Interpreting the functional impact of genetic variations is crucial for understanding diseases and traits, especially with the explosion of whole-genome sequencing (WGS) data. While previous tools like the original FAVOR existed, they had limitations: they only provided global annotations (not tissue-specific), lacked gene/protein-level details, had limited visualization, and only supported one reference genome (hg38). This created a bottleneck for researchers trying to make sense of the vast amount of genomic information generated by large-scale studies like GWAS and WGS. FAVOR 2.0 aims to overcome these limitations by providing a more comprehensive, user-friendly, and versatile platform.
Methods
FAVOR 2.0 is built on a completely redesigned infrastructure. Key methodological advancements include:
- FAVOR-API: A new, high-performance, scalable API that powers the entire platform and allows programmatic access for other tools and pipelines.
- Revamped Web Interface: Built with Next.js for a responsive and interactive user experience, supporting both desktop and mobile devices.
- Interactive Genome Browser: Utilizes libraries like Gosling and HiGlass to visualize multi-faceted variant annotations across genomic regions with multiple data tracks.
- Network Visualization Modules: Employing Cytoscape.js to visualize protein-protein interactions (PPIs) and pathway networks.
- FAVOR-GPT: An AI-powered chatbot leveraging a retrieval-augmented generation (RAG) framework with multiple LLMs (GPT 5 nano, GPT 40 min, GPT 4.1 nano, DeepSeek R1) to enable natural language queries and provide interpretable explanations of genomic data.
- Dual Reference Genome Support: Building separate databases for both hg38 and hg19 reference genomes from scratch to ensure accuracy and compatibility with legacy data.
Datasets
FAVOR 2.0 integrates data from numerous sources, which are updated annually or as new versions are released. Key sources mentioned include:
- gnomAD: For allele frequencies.
- ClinVar: For clinical significance of variants.
- GTEx: For tissue-specific gene expression profiles.
- ENCODE: For candidate cis-regulatory elements (CCREs).
- MPRA data: For high-throughput measurements.
- eQTL datasets: For expression quantitative trait loci.
- PPI databases: BioGRID, IntAct, HuRI.
- Pathway databases: KEGG, WikiPathways, BioCyc, IntPath.
- Other sources: Mutation Rate, SpliceAI, COSMIC, GWAS Catalog, Gnocchi, JARVIS, AlphaMissense, MACIE, SCENT, pgBoost, CATlas, CRISPR screens.
The paper does not introduce new benchmark datasets but rather integrates and enhances existing ones.
Key Insights and Ideas
- Democratizing Genomic Interpretation: The integration of FAVOR-GPT is a significant step towards making complex genomic data accessible to a wider audience, including those without deep bioinformatics expertise, through natural language queries.
- Bridging Legacy and Modern Data: Providing support for both hg38 and hg19 reference genomes is crucial for integrating new WGS data with existing GWAS and clinical datasets, avoiding potential errors from coordinate conversion.
- Multi-Omics Integration: FAVOR 2.0 moves beyond simple variant annotation by integrating gene, protein, and epigenomic data, offering a more holistic view of functional impacts.
- Enhanced Visualization for Discovery: The interactive genome browser and network visualization tools are powerful for hypothesis generation and exploring complex relationships between variants, genes, and pathways.
- Scalability and Accessibility: The new API and cloud-based infrastructure ensure the platform can handle large datasets and remain sustainable for long-term use.
Key Results
- Expanded Annotations: FAVOR 2.0 offers dramatically expanded annotations, including global, tissue- and cell-type-specific variant annotations, as well as gene and protein annotations.
- Dual Genome Support: Comprehensive support for both hg38 and hg19 reference genomes is now available.
- FAVOR-GPT Performance: The AI chatbot has been validated to ground its responses in the FAVOR database, minimizing hallucinations and ensuring high fidelity of query results.
- Scalable API: The FAVOR-API supports queries of up to five million variants per request, enabling efficient analysis of large inputs.
- User Experience: The revamped web interface is described as more user-friendly and interactive, with faster page loads and fluid navigation.
Limitations and Future Work
- Limitations: The paper acknowledges that FAVOR-GPT, as an AI-assisted system, may occasionally generate incomplete or inaccurate results due to the probabilistic nature of language models. Users are advised to interpret outputs critically and verify findings.
- Future Work: The authors plan to continue expanding and improving the quality and scope of functional annotations. They aim to integrate emerging experimental, genetic, genomic, and phenotypic data. A key future direction is the development of "FAVOR-agent," an advanced evolution of FAVOR-GPT designed for multimodal data integration (genomic, transcriptomic, epigenomic, proteomic) and natural language-driven data manipulation and workflow execution (e.g., VCF file annotation, genotype data conversion). They also plan to support multiple languages for FAVOR-GPT to break down language barriers.
Summary updated