ChatMDV: reducing technical barriers in bioinformatics analysis using large language models

The rapid advancement in single-cell, spatial omics, imaging, and genomic technologies requires robust analytical and visualisation platforms capable of managing complex biological data. Tools such as Multi-Dimensional Viewer (MDV) offer comprehensive interfaces for data exploration but often require advanced computational expertise and manual configuration to generate visualisation outputs, limiting accessibility for many users.

We present ChatMDV, a natural language interface integrated with MDV that enables users to generate high-quality, interactive visualisations and analyses through natural language commands. ChatMDV employs a retrieval-augmented generation pipeline in combination with large language models to translate user queries into executable, reproducible Python code and interactive output. This conversational layer facilitates both exploratory and targeted analyses in diverse biological domains. We demonstrate ChatMDV's capabilities using 3 datasets of increasing complexity: the Peripheral Blood Mononuclear Cells 3K single-cell RNA-sequencing (scRNA-seq) dataset, the lung cancer atlas scRNA-seq dataset included in the Human Cell Atlas, and the longitudinal TAURUS study scRNA-seq dataset. Across all use cases, ChatMDV produced high-quality, reproducible visualisations from simple natural language queries, achieving a high semantic success rate between 79% and 97% when visualising the datasets.

By bridging the gap between natural language processing and bioinformatics visualisation, ChatMDV reduces technical barriers, enhances reproducibility, and supports more inclusive scientific inquiry. Its modular design and adherence to Findability, Accessibility, Interoperability, and Reuse (FAIR) principles make it a scalable and adaptable framework for accelerating biological data analysis.