BITTENSOR · SUBNET 107 · WHITEPAPER

DNA Mutation
Inference &
Benchmarking

A unified overview of the Minos architecture, a Bittensor subnet that uses decentralized, incentive-driven optimization to solve the variant calling problem in genomics.

Covers HelixForge synthetic mutation injection, blind evaluation framework, multi-component scoring, EMA smoothing, Winner-Take-All incentive design, and the five-phase roadmap from hyperparameter optimization to production deployment.

Pages20
Versionv1.1
DateFeb 2026

Document Details

Minos Overview

VIEW PDF ↗

Decentralized Infrastructure for DNA Mutation Inference & Benchmarking: architecture, incentive mechanism, scoring, security properties, and the road ahead.

As DNA sequencing costs have fallen from $1 million to under $200 per genome, the bottleneck in genomics has shifted from data generation to data analysis. Variant calling, the process of identifying DNA mutations from sequencing data, is the critical inference layer, and despite major investment from Google, the Broad Institute, and government agencies, no single tool is consistently the most accurate. Minos is a Bittensor subnet that addresses this through decentralized, incentive-driven optimization. Every 72 minutes, the platform generates a fresh challenge genome containing hidden synthetic mutations injected at the read level. Miners optimize variant calling tool configurations, and in later phases, submit entirely custom algorithms, while validators execute them in isolated containers and score the results against known ground truth. The best-performing miner receives all the emissions.

Key Insight

This continuous benchmarking produces three valuable outcomes. First, as a byproduct, Minos builds the world's largest validated synthetic genome database: over thousands of validated genomes in only the first year, each with confirmed ground-truth mutations. Second, by aggregating the top-performing models across tools and genomic contexts, Minos trains a consensus variant caller expected to exceed the accuracy of any individual method. Third, this consensus model will be deployed as AI-powered variant calling infrastructure where hospitals, biobanks, and pharmaceutical companies can submit sequencing samples and receive high-accuracy variant calls, reducing missing diagnoses and minimizing clinical errors through continuously validated, decentralized AI algorithms developed by Minos.

Details
20 PAGESV1.1FEBRUARY 2026