omnistrand

## Introduction: A Humble Experiment

I went into mva hackathon with little to no background in genetics. To me, a genome was just a massive text file. I wanted to build an experimental tool to bridge the gap between finding a 1D genomic mutation in a VCF file and visualizing its 3D structural impact, without needing a PhD to operate it.

I built OmniStrand—a React and TypeScript web app that parses patient VCF files 100% locally in the browser for absolute privacy. To make it accessible, I integrated a custom WebMCP (Model Context Protocol) architecture. This allows an autonomous LLM agent to literally pilot the user interface—automatically typing in coordinates, navigating the genome browser, and pulling up the 3D AlphaFold structures via PDBe-Molstar.

## The Submission: A Perfect Score

I used this experimental tool as my actual pipeline for the hackathon's clinical benchmark. I loaded the patient VCF into OmniStrand, and the WebMCP agent took over the UI. It successfully navigated the genome, identified the compound heterozygous mutation responsible for Mosaic Variegated Aneuploidy (MVA), resolved the UniProt ID, and highlighted the mutated residue on the 3D protein model.

The AI then generated my clinical CSV payload directly from the UI. Despite my lack of prior genetics experience, I submitted that exact payload for Track 1, and it was accepted with a perfect score.

## Future Improvements

While OmniStrand successfully automated a clinical workflow, this is just the foundation. My next steps for the project include:

• Multi-Agent Orchestration: Expanding the WebMCP toolset so multiple AI agents can collaborate in the UI (e.g., one agent parses the genome while another simultaneously queries pharmacological databases for drug interactions). • In-Browser Molecular Docking: Upgrading the 3D Discovery Engine to support basic ligand-docking simulations, allowing researchers to test drug repurposing directly in the browser. • Wasm-Powered Genomics: Replacing the JavaScript VCF parser with a WebAssembly (Wasm) implementation to handle massive whole-genome sequencing (WGS) files with zero lag.