DeSci-AI (Subnet TBD) — Subnet Design Proposal
Ideathon Submission · Bittensor Subnet Hackathon · 2026
| Subnet Name | DeSci |
| Tagline | AI-Native Drug Discovery on Bittensor |
| Status | Phase 0 PoC — AlphaFold-2 peptide generation pipeline implemented |
| Backend Repo | DeSci/ — Bittensor miner, validator, Platform API |
| Frontend | DeSci-web/ — Landing page, architecture visualization, API demo |
| Demo API | POST /api/v1/peptides/generate (FastAPI, local / Docker) |
| Documentation | README.md, interactive OpenAPI at /docs |
DeSci-AI is a Bittensor subnet designed to democratize pharmaceutical R&D compute and accelerate therapeutic discovery through decentralized GPU infrastructure and autonomous AI agent pipelines.
Our core vision is to transform how drug discovery workloads are executed: instead of pharma companies renting centralized cloud GPU clusters at $50K–$500K/year, researchers submit structure prediction, molecular docking, and peptide generation tasks through a REST API gateway. Distributed miners compete to deliver the highest-quality computational biology outputs — validated objectively through pLDDT scores, docking affinities, and ADMET predictions — and earn TAO proportional to output quality via Yuma Consensus.
The mined commodity is validated drug discovery compute — peptide sequences with AlphaFold-2 structure predictions, molecular docking scores, and eventually full multi-step discovery pipeline outputs.
DeSci-AI operates as a dual-engine architecture:
| Engine | Audience | Value Proposition |
|---|---|---|
| 2B — Compute-as-a-Service | Pharma enterprises, biotechs, CROs | REST API gateway to supercomputing-grade GPU resources at 60–70% of AWS/GCP cost; TEE-protected confidential computations |
| 2P — Agent-Driven Pipelines | Internal R&D + licensing partners | Multi-agent swarms (Disease Strategist → Target Discovery → Molecular Designer → ADMET) progressing autonomously from hypothesis to Preclinical Candidate (PCC) nomination |
Phase 0 (current PoC) implements the 2B engine's first workload: AlphaFold-2 peptide generation against target proteins. The full roadmap extends to 30+ integrated open-source models across six categories — structure prediction, molecular docking, molecular generation, ADMET prediction, MD simulation, and retrosynthesis — orchestrated through a unified workflow DSL.
This proposal outlines the subnet's incentive structure, miner and validator roles, market rationale, and system architecture, demonstrating how DeSci-AI represents a genuine Proof of Intelligence for computational drug discovery.
The DeSci-AI incentive mechanism aligns miners, validators, and enterprise users toward a shared goal: producing verifiably high-quality drug discovery computations at lower cost than centralized alternatives.
DeSci-AI follows the standard Bittensor emission split:
Total Alpha Emissions per Tempo
├── 41% → Miners (incentive-weighted by composite score)
├── 41% → Validators + Stakers
└── 18% → Subnet Owner (Platform API + orchestration infrastructure)
Miner rewards are proportional to composite quality score W_i via Yuma Consensus:
R_i = (Δα × 0.41) × (W_i / Σ W_j)