StateZero.Labs

ENGINE::CORE_COMPUTE

Core Compute Engine

A three-stage pipeline that translates single-cell transcriptomic disease coordinates into synthesizable small-molecule candidates — entirely in silico.

LINCS Screening──▶Virtual CRISPR──▶TFE Encoder──▶Candidate

STAGE::01

Biological Ligand Screening

Screens the NIH LINCS Phase I L1000 dataset (GSE92742) to evaluate recombinant ligands, cytokines, and morphogens. The engine calculates multi-dimensional cell-state transitions using Cosine Similarity, Earth Mover's Distance (EMD), and Mean Squared Error (MSE) to identify vectors that collapse the fibrotic manifold into the regenerative fetal blueprint.

GSE92742trt_lig filterCosine SimEMDMSE
LigandCos_Sim (Trajectory)EMD (Manifold Shift)
EGF ●0.921.14
FGF20.851.42
TGFB1 ✕−0.655.89

STAGE::02

Virtual CRISPR RL Agent

Reinforcement learning simulates parallelised multi-gene knockouts targeting scale-free master regulatory hubs — collapsing fibrotic trajectories via TGF-β / SMAD pathway suppression.

Simulation running
Knockouts evaluated0
Active targetTGF-β / SMAD3
Network scan0%

> hub collapse: SUCCESS

> trajectory shift: −0.83σ

> next iteration...

STAGE::03

Transcriptomic-to-Fingerprint Encoder

Translates the target genetic coordinate vector — the output of the Virtual CRISPR agent — directly into Morgan Fingerprint chemical space. Maps biological reprogramming intent to synthesizable polypharmacology candidates without manual chemical intuition.

Encoding Pipeline

ΔExpression Vector──▶Latent Embedding
Latent Embedding──▶Morgan Fingerprint
Morgan Fingerprint──▶Candidate Molecule
Morgan FP2048-bit vectorTanimoto similarityPolypharmacology
StateZero Labs © 2026