Preprint · 2026-05-22 · Pre-submission draft
Manuscript, IP, BD — all open.
bioRxiv v0.2 Methods + Discussion drafts, KIPO 5-patent follow-up outline (ROBOGATE), 5 manuscript figures, plus archived exploratory TNIK materials (superseded — see Section 4). All under DigitalBiology Co., Ltd. authorship — patent series 10-2026-0057732 under 18-month confidentiality until 2027-09.
Section 1 · Manuscript
bioRxiv v0.2 drafts
Section 2.1 – 2.11
bioRxiv v0.2 — Methods (full)
Saturn Mamba SSM config, Boltz-2 / Chai-1 / Protenix ensemble, AEV-PLIG v3a, GNINA orthogonality, Vina, AiZynthFinder, Mondrian Conformal Prediction, 7-axis GOLD filter, hardware + reproducibility, ROBOGATE failure-boundary.
Download Markdown →
Section 4 – 10
bioRxiv v0.2 — Discussion + Conclusion
Multi-axis filter hit-rate hypothesis (CACHE 3.7% baseline), GNINA-AEV orthogonality novelty, MolFormer dataset-size threshold, Insilico Rentosertib IP comparison, 7-axis Tier 1 validation, 6 limitations disclosed, R2 q=5 portfolio recommendation.
Download Markdown →
Pre-deploy structure
bioRxiv v0.1 — Original outline
Earlier outline used during platform stabilization.
Download Markdown →
Section 2 · Manuscript figures
5 figures · PNG · open license
Section 3 · IP & licensing
Patents + licensing narrative
ROBOGATE failure-boundary heatmap
KIPO 10-2026-0057732 series — 5 follow-up patents
Filed parent 10-2026-0057732 (18-month confidentiality until 2027-09). 5 follow-up outlines: adaptive sampling, multi-axis GOLD scoring, GNINA-AEV orthogonal consensus, kinome selectivity anchor, in silico ROBOGATE heatmap encoding.
Download Markdown →
Rentosertib-anchored · superseded
TNIK IP-novelty analysis (archived / exploratory)
ARCHIVED. Exploratory IP-novelty comparison of MolForge TNIK chemotypes vs Insilico Rentosertib. The TNIK program is deprioritized/paused and its former lead was dropped at synthesis — this document is retained as history only, not as a licensing offer.
Download Markdown →
Section 4 · Archived exploration
TNIK BD materials — archived / superseded
TNIK · superseded · not a licensing offer
R2 CRO deck (markdown) — archived exploratory draft
ARCHIVED. Historical R2 evidence-deck draft from the (now deprioritized) TNIK exploration. The candidate it centred on was dropped at the synthesis stage — this is retained as process history only, not as a licensable lead.
Download Markdown →
DigitalBiology → CRO partner · superseded
R2 CRO outreach email — archived template
ARCHIVED template from the TNIK exploration. Not used for active outreach; retained for internal reference only.
Download Markdown →
Authors
Heonjeong Cho (Co-CEO, conceptualization · scientific direction) · Jewoo Yom (CTO, computational infrastructure · model development) · DigitalBiology Co., Ltd. (Seoul, 752-88-03445).
Status
Pre-submission draft. R1 wet-lab results have returned: three TYK2 candidates measured at ~2 nM potency, confirming binding to the JAK family. Note that our historical “TYK2” data was seeded from CHEMBL2148, which corresponds to JAK3 (real TYK2 = CHEMBL3553); the R1 hits are validated JAK-family / pan-JAK binders that cross-bind TYK2, and TYK2-selectivity is not yet proven. The R1 mechanism observed was ATP-competitive (JH1 catalytic site), not allosteric JH2. Conformal recalibration on the small (n=3) R1 set is provisional. Numbers in the archived draft docs predate these corrections.
Conflicts of interest
Authors are co-founders and equity holders in DigitalBiology Co., Ltd. The ROBOGATE method is covered by the KIPO 10-2026-0057732 patent series. No individual compound is presented here as a licensable lead.
License
Manuscript drafts under CC-BY 4.0. Figures and code under MIT. Patent series under 18-month KIPO confidentiality until 2027-09 — outlines published here describe scope only, not implementation details.