CITATION.cff

cff-version: 1.2.0
message: "If you use this software, please cite it as below."
title: "Darwin PBPK Platform: AI-Powered Pharmacokinetic Prediction with Validated Dataset"
version: 1.0.1
date-released: 2025-11-08
doi: 10.5281/zenodo.17536674
repository-code: "https://github.com/agourakis82/darwin-pbpk-platform"
license: MIT
type: software

authors:
  - family-names: "Agourakis"
    given-names: "Demetrios Chiuratto"
    email: "demetrios@agourakis.med.br"
    affiliation: "Pontifical Catholic University of São Paulo (PUC-SP); Faculdade São Leopoldo Mandic"
    orcid: "https://orcid.org/0000-0002-8596-5097"

abstract: >
  Darwin PBPK Platform v1.0.1 is a state-of-the-art deep learning system for
  predicting physiologically-based pharmacokinetic (PBPK) parameters
  using multi-modal molecular representations. NEW in v1.0.1: Validated PBPK
  dataset publication with complete metadata, experimental validation protocols,
  and quality assurance documentation. The platform integrates ChemBERTa embeddings
  (768d), molecular graphs via PyTorch Geometric (20 node + 7 edge features), and
  RDKit molecular descriptors (25 features) to predict fraction unbound (Fu),
  volume of distribution (Vd), and clearance (CL) with high accuracy (target R² > 0.55).
  Trained on 44,779 compounds from ChEMBL and Therapeutics Data Commons (TDC), the
  system employs advanced GNN architectures (GAT + TransformerConv with 4 attention
  heads each) with multi-task learning and physics-informed PhysioQM constraints.
  Includes dataset documentation, validation metrics, and reproducibility protocols
  for drug discovery pipelines and Q1 publication in computational chemistry and
  machine learning journals (Nature Machine Intelligence, JAMA, Journal of Medicinal Chemistry).

keywords:
  - PBPK
  - pharmacokinetics
  - drug discovery
  - deep learning
  - graph neural networks
  - ChemBERTa
  - multi-task learning
  - ADMET
  - computational chemistry
  - PyTorch Geometric
  - dataset publication
  - validation protocols