Features
ExpressionVAE discrete-token encoding for single-cell data
Clinical response prediction from baseline biopsies
Virtual cell models encoding patient biology
Pre-clinical to clinical data integration
Trial design optimization (e.g., power analysis)
Drug rescue identification for failed compounds
Foundation models trained on multi-omics data
Presentation at ICLR '26, CSHL '26, ICML '26
Public research blog with technical deep dives
Foundation model vBx-1.0 for precision neurology
Multimodal patient brain dataset: 12,000+ brains, 6,000 patients
Paired proteomic, genomic, and clinical data per patient
Physical inventory of 900+ frozen brain tissue samples
Virtual biopsy model predicts brain activity from blood draw
Target discovery from proprietary human data
Target characterization with 83% preclinical validation rate
Biomarker identification for clinical trials
Patient stratification for trial enrichment
Lifts responder fraction from 52% to 69% (vBx-1.0)
Reduces trial enrollment by 43% in Parkinson's (L-DOPA responders)
Platform for target discovery, characterization, biomarkers, and stratification
Supports Parkinson's, Alzheimer's, and other neurodegenerative diseases
Partnership model with top-20 pharma collaborations
