Jove
Visualize
Contact Us
JoVE
x logofacebook logolinkedin logoyoutube logo
ABOUT JoVE
OverviewLeadershipBlogJoVE Help Center
AUTHORS
Publishing ProcessEditorial BoardScope & PoliciesPeer ReviewFAQSubmit
LIBRARIANS
TestimonialsSubscriptionsAccessResourcesLibrary Advisory BoardFAQ
RESEARCH
JoVE JournalMethods CollectionsJoVE Encyclopedia of ExperimentsArchive
EDUCATION
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab ManualFaculty Resource CenterFaculty Site
Terms & Conditions of Use
Privacy Policy
Policies

Related Experiment Videos

Developing Predictive Models by Sharing Predictions - An Investigation of a Federated Learning Approach for ADMET

Rajarshi Guha1, Wenyi Wang2, Edward Price3

  • 1Vertex Pharmaceuticals, 50 Northern Avenue, Boston, Massachusetts 02129, United States.

Journal of Medicinal Chemistry
|May 29, 2026
PubMed

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by
Same author

Preclinical characterization of EGT710, an oral non-peptidomimetic reversible covalent SARS-CoV-2 main protease inhibitor.

npj drug discovery·2026
Same author

Computational Identification of Active Drug Metabolites for Human Protein Targets.

Molecular pharmaceutics·2026
Same author

Optimization and <i>In Vivo</i> Characterization of a Series of Cbl‑b Inactive-State Inhibitors.

ACS medicinal chemistry letters·2026
Same author

Loss of the autoimmune risk gene TREX1 reveals a convergence of mechanisms promoting immune tolerance loss and antitumor immunity.

Science advances·2026
Same author

Evaluation of Oral PROTAC Guidelines: Efflux Ratio Outweighs Chameleonicity Descriptors.

ACS medicinal chemistry letters·2026
Same author

Automated Molecular Design in BRADSHAW, Applied to the Optimization of ERAP1 Inhibitors.

Journal of medicinal chemistry·2026
Summary
This summary is machine-generated.

Pharmaceutical companies can securely develop machine learning models for drug discovery using a novel student-teacher model (STM) framework. This approach enables collaborative ADMET prediction without sharing proprietary data, enhancing model development.

Area of Science:

  • Computational chemistry
  • Pharmacology
  • Artificial intelligence in drug discovery

Background:

  • ADMET prediction models require large, diverse datasets, which are often siloed across organizations.
  • Federated learning (FL) offers a solution for collaborative modeling while maintaining data privacy.

Purpose of the Study:

  • To investigate a student-teacher model (STM) framework for secure, collaborative development of ADMET prediction models.
  • To assess the feasibility of STM using predictions from multiple pharmaceutical companies.

Main Methods:

  • Organizations trained internal machine learning models on proprietary data.
  • Predictions from these models were used to generate pseudolabels on a public dataset.
  • A centralized student model was trained on this pseudolabeled dataset.

Related Experiment Videos

Main Results:

  • The student model achieved performance comparable to individual teacher models (RMSE ≈ 0.51).
  • A pseudolabeled dataset of approximately 133,000 compounds was generated for rat steady-state volume of distribution prediction.
  • The STM framework demonstrated practical scalability for cross-company model development.

Conclusions:

  • The student-teacher model (STM) provides a simplified and secure alternative to existing federated learning approaches.
  • STM facilitates collaborative ADMET prediction model development without direct data sharing or iterative collaboration.
  • This framework offers a practical and scalable solution for pharmaceutical companies to enhance drug discovery through secure, cross-company AI model building.