Drug developers with approvals and exits behind them. Bioinformaticians. AI data scientists. Engineers who ship. Not a software team that hired a chemist, and not a pharma team that hired a data scientist — one group running all three layers at once.
Taking an approved active, re-engineering how it is delivered, and carrying it through approval and exit. That is the exact model behind SAPU003, and this team has already run it end to end.
Albumin-bound paclitaxel. FDA approved January 2005 in metastatic breast cancer, later non-small cell lung and pancreatic. Abraxis BioScience acquired by Celgene for $2.9 billion in 2010.
Paclitaxel in a diblock copolymer micelle, marketed as Genexol-PM. Sold by Sorrento Therapeutics to NantPharma for up to $1.3 billion.
Abraxane and Cynviloq went after tumours everyone already knew about — the science was in the delivery. SAPU003 goes after a driver the engine surfaced, in a field nobody was formulating for. The formulation half is the part these people have already proven twice. The target half is what's new.
Most companies in this category are one of these wearing another's coat. The gap between a ranked gene list and an IND is where they stall.
Formulation, GMP manufacture, IND-enabling packages, regulatory strategy. Approvals and exits already on the record — not advisors, operators.
Survival analysis, cohort work, biomarker validation. The job is to try to kill a candidate on public patient data before anyone spends money on it.
Data scientists and engineers who took 28M abstracts into production on a vector layer, and keep it refreshed. The method is theirs — patent-pending, favorable WIPO ISR.
SAPU003 and SAPU006 were not internal science projects. Sapu Nano engaged the team and funded the work — concept to clinic in 24 months for $1–3M, twice. Someone with a budget had already decided the engine was worth paying for before it was ever a company.
PDAOAI holds SAPU003 in pediatric neurology and longevity, in-licensed from Sapu Biosciences. Sapu retains worldwide oncology and the compound IP. That is the model in one sentence: run the engine, prove it on someone else's dime, take a field position in what it finds.