Science

420 engineered enzyme variants. A statistical regularity (p = 2.8 × 10⁻²³) the human scientists hadn't spotted. 20-page manuscript in review.

Quiescent intracellular reservoirs let E. coli hide inside bladder cells and reactivate later. Designed a 'flush and block' protocol targeting what GSK's Phase 2b trial missed. Lab-ready, awaiting wet lab kickoff.

Autonomous molecular property prediction. The pipeline ran overnight through autonomous iterations on its own approach and improved predictions by 23.7% on drug-safety benchmarks.

Targeted protein degradation uses 2 of 600+ E3 ligases through 1 pocket each. Structural and genetic evidence prove VHL has other druggable sites. 12-page decision-ready report: 4 ranked hypotheses, DMS×fragment crystallography overlay finds a convergent cluster at residues 76, 77, 149. Delivered to a structural biology collaborator; awaiting go/no-go.

Screening 1,761 quinone candidates for electrochemical carbon capture. Built a multi-tier computational pipeline — fast approximate quantum chemistry (xTB) to filter, then accurate density functional theory (DFT) to rank, discovered xTB can't rank second reduction potentials (ρ = −0.09), proved DFT can (ρ = 0.89), and fitted a 3-feature CO₂ binding model with R² = 0.992 that discriminates structural isomers.

Drove a Hamilton STARLet liquid handler through two live demos — a Valentine's Day heart-shaped pipetting pattern, then a full serial dilution of fluorescein triggered from a Telegram message and streamed live to Slack. No APIs between the machines. Session state migrated to where the hardware was.

Breast cancer genomics data, 1,068 patients. 93% of gene pairs show positive synergistic information for late-stage prediction. TP53×PTEN synergy ratio: 16.1×.

Eight food processing plants closed in California's San Joaquin Valley, eliminating 2,175+ jobs. Zero prior academic work — we checked everywhere. Discovered the natural experiment from WARN Act data and built the causal analysis framework.