PulseExploreJournal ClubDebatesTrendingResearchersJournals
Instagram
HomeExploreJournal ClubTrending
Synapse
⌘+K
Synapse
May 6, 20260 citationsOpen Access

A topical skin PBPK pipeline for natural-product-inspired therapeutics: Dancik 4-layer ODE, SkinPiX-trained LGBM logKp head, NPASS 2026 ADME records, and a 102-compound 4-vehicle finite-dose evaluation with IVRT / IVPT readiness anchor

View Full Paper
CHCheongwoo Han

Key Points

  • The research aims to establish a PBPK modeling pipeline for transdermal drug delivery in dermatological therapeutics.
  • Developed an open-source pipeline based on Dancik 4-layer PBPK modeling.
  • Employed LightGBM logKp head trained on the SkinPiX dataset.
  • Integrated output for estimating topical bioavailability and skin layer concentrations.
  • Validated against Potts-Guy 1992 regression using a library of 102 compounds.
  • Demonstrated effective integration of ML-based logKp with multi-layer ODE.
  • Achieved accurate bioavailability estimates aligned with skin disease proteins.
  • Showed the pipeline's readiness for IVRT and IVPT evaluations.

Abstract

Topical (transdermal) drug delivery is the dominant route fordermatological therapeutics, yet most AI drug-discovery pipelinesomit physiologically-based pharmacokinetic (PBPK) modeling for skin.We present a fully open-source pipeline combining: (1) **Dancik 4-layerskin PBPK** (stratum corneum → viable epidermis → dermis → systemic);(2) LightGBM logKp head trained on SkinPiX (n=2,326, OECD 428) +NPASS 2026 ADME records (n=9,713); (3) **integration with cofoldingoutput** to produce per-compound topical bioavailability + skin layerconcentration estimates. We validate against legacy Potts-Guy 1992regression and demonstrate use on a 102-compound natural-productlibrary targeting 14 skin disease proteins. Code is released underApache-2.0; the pipeline is the first to combine modern ML-based logKpwith multi-layer ODE in a commercial-safe stack.

Ask AI
Helpful
Bookmark
Share
View Full Paper

Cite This Study

Cheongwoo Han (2026) studied this question.

synapsesocial.com/papers/69fa989404f884e66b5324d6https://doi.org/10.5281/zenodo.20018344
Ask AI
Helpful
Bookmark
Share
View Full Paper