I. Open-Source Molecular Discovery
Bypassing the Proprietary Bottleneck.
Traditional pharmaceutical research relies on highly guarded, proprietary infrastructure that fundamentally slows the pace of discovery. At 33med, we leverage the power of open-source computational biology and custom LLM automation to research Ankylosing Spondylitis (AS) with complete autonomy.
We are systematically mapping the cellular dysfunctions that drive AS—specifically targeting the HLA-B27 receptor and Ankylosing Spondylitis Mesenchymal Stem Cells (AS-MSCs). By engineering our own visualization and simulation stack, we test therapeutic interventions at the speed of software.
II. Phase 1 – Single-Cell Analysis & Target Identification
Pinpointing the Genetic Misfires.
Before we can computationally simulate a cure, we must map the exact biological errors. Our pipeline ingests massive public RNA sequencing datasets to isolate specific genetic misfires, such as BMP2 overexpression, occurring within AS-specific stem and immune cells.
Our Target Identification Stack:
Scanpy & Seurat
The core computational engines utilized for filtering, clustering, and performing high-speed differential expression analysis on massive single-cell RNA-seq (scRNA-seq) datasets.
Bioconductor
An advanced R-based suite deployed for rigorous genomic data preprocessing and normalization.
StemChecker
A targeted computational tool used to rapidly verify cell identity against curated stemness genetic signatures.
ceLLama & scExtract
LLM-integrated automation pipelines that empower our AI agents to rapidly and autonomously categorize and annotate newly published single-cell data.
III. Phase 2 – Molecular Simulation & Intervention Testing
Testing Therapies in a Digital Sandbox.
Once we identify a genetic target, we transition from analysis to active simulation. We generate precise 3D structures of target proteins, simulate their physical interactions in fluid environments, and predict exactly how AS-MSCs will respond to specific drugs or genetic edits—entirely in silico.
Our Simulation & Testing Stack:
AlphaFold & ESMFold
AI inference models that predict the exact 3D folded structure of complex proteins, including the HLA-B27 receptor, directly from raw amino acid sequences.
GROMACS & OpenMM
Heavily GPU-accelerated molecular dynamics (MD) engines used to simulate the physical movement and atomic forces of proteins over time.
AutoDock Vina & RDKit
Foundational cheminformatics toolkits and high-throughput docking engines used to calculate binding affinities—verifying if a proposed molecular inhibitor successfully attaches to the target receptor.
Pertpy
A specialized Python simulator used to test single-cell perturbations, allowing us to model the exact reaction of AS-MSCs when specific genes are computationally triggered or suppressed.