OpenCure Labs

OpenCure Labs

GitHub

About OpenCure Labs

Open infrastructure for computational biology โ€” with a long-term goal of enabling more open and accessible personalized medicine workflows.

Mission

OpenCure Labs exists to make computational biology research more open, reproducible, and accessible. We run autonomous AI-driven research pipelines across genomics, drug discovery, and protein science โ€” then publish every result as a signed, inspectable artifact that anyone can download and verify.

The platform generates research tasks from public biomedical datasets (TCGA, ClinVar, ChEMBL, GEO), executes them on GPU compute via specialist agents, and publishes the outputs with full provenance metadata. Every artifact carries an Ed25519 integrity signature for auditability.

How the System Works

๐Ÿงฌ

1. Task Generation

Research tasks derived from public biomedical datasets exploring open problems in oncology, rare disease, and pharmacology.

๐Ÿค–

2. Agent Execution

Specialist agents run computational pipelines on GPU clusters โ€” neoantigen binding, variant scoring, molecular docking, QSAR.

๐Ÿ”ฌ

3. Two-Tier Grok Review

Grok independently critiques each result: methodology, confidence scoring, and literature overlap. Below-threshold results are blocked.

๐ŸŒ

4. Signed Artifact Publication

Approved artifacts stored in Cloudflare R2/D1 with Ed25519 signatures, provenance metadata, and validation warnings.

Research Skills

๐ŸŽฏ

Neoantigen Prediction

Tumor-specific peptide-MHC binding for personalized cancer immunotherapy targets.

๐Ÿงฌ

Variant Pathogenicity

Genetic variant scoring for disease-causing potential using ensemble ML models.

โš—๏ธ

Molecular Docking

Protein-ligand interaction simulation to identify drug binding candidates.

๐Ÿ“Š

QSAR Modeling

Structure-activity models predicting compound bioactivity from molecular features.

๐Ÿ”ฌ

Structure Prediction

Protein 3D structure prediction for downstream docking and analysis.

๐Ÿ“‹

Sequencing QC

Quality control โ€” coverage, error rates, and contamination checks.

๐Ÿ“š

Literature Review

AI-powered analysis of scientific literature to identify gaps and validate hypotheses.

The Review Process

Every result passes through a two-tier review before publication:

Tier 1: Automated Validation

A local critique step checks methodology, input data quality, and statistical validity. Results that fail local validation are withheld.

Tier 2: Independent Critique (Grok)

Grok (xAI) independently evaluates each result โ€” critiquing methodology, checking literature overlap, and assigning a confidence-oriented review score. Below-threshold results are blocked with a written explanation.

Multi-Species Support

๐Ÿงฌ

Human

HLA-based neoantigen prediction, variant pathogenicity, and drug response across TCGA cancer types.

๐Ÿ•

Dog (Canine)

DLA-88 MHC allele support for canine neoantigen prediction and variant analysis.

๐Ÿˆ

Cat (Feline)

FLA MHC allele support for feline neoantigen prediction and variant analysis.

Why the Public Artifact Browser?

We built the public browser at opencurelabs.ai because open science means more than open-source code โ€” it means open data. Every computational result we generate is published with integrity signatures, provenance metadata, and explicit validation warnings so anyone can inspect, reproduce, and build on the work.

With 25,000+ published artifacts and growing, the browser gives researchers instant access to AI-generated hypotheses across oncology, rare disease, and drug discovery โ€” no account required.

Responsible AI Disclaimer

All outputs are AI-generated computational predictions. They carry explicit validation warnings and have not been experimentally validated. Each artifact includes provenance metadata and an Ed25519 integrity signature for auditability. Outputs are open for inspection and reuse, but should be independently validated before scientific or clinical use.

Open Source

The full platform โ€” agents, pipelines, coordinator, reviewer โ€” is open source on GitHub. Contributions are welcome via pull requests.

Built by Shone Anstey

Made with ๐Ÿ”ฌ in Vancouver ๐Ÿ‡จ๐Ÿ‡ฆ

@shoneanstey ยท GitHub