Sign in Subscribe
Machine learning

Machine Learning Models Predict Steroidogenesis with High Accuracy

Strategic Insights

— 1 min read

Scientists at Collaborations Pharmaceuticals Inc. have developed advanced machine learning models that can accurately predict the impact of chemicals on steroidogenesis, a critical process in the human body that regulates hormone production. The research, published in Environmental Science & Technology in 2025, employed high-throughput screening and computational models to identify effective steroidogenesis modulation models. The models were validated using a prospective test set of 20 compounds, achieving 80% accuracy with conformal prediction adjustments.

Key Takeaways:

  • The researchers developed a random forest model that predicted steroidogenesis inhibition with 80% accuracy using a prospective test set of 20 compounds.
  • A transformer-based model (MolBART) was also developed to predict all end points simultaneously, achieving consistent performance.
  • The models were trained on data from approximately 1,800 chemicals screened in H295R human adrenocortical carcinoma cells.
  • The models enable predictions of both general steroidogenesis inhibition and potential molecular targets.
  • The research provides a rapid and scalable system for assessing chemical impacts on steroidogenesis, supporting chemical risk assessment, product stewardship, and regulatory decision-making.
  • The study was funded by the NIH National Institute of General Medical Sciences (NIGMS) and NIH National Institute of Environmental Health Sciences (NIEHS).
  • Additional authors include Patricia A. Vignaux, Joshua S. Harris, Scott H. Snyder, Fabio Urbina, and Sean Ekins.
  • The models were developed using data from ChEMBL, a publicly available database of bioactive compounds.

Statistics:

  • The research employed high-throughput screening of approximately 1,800 chemicals.
  • The random forest model achieved 80% accuracy with conformal prediction adjustments.
  • The transformer-based model (MolBART) was validated using a test set of 20 compounds.
  • The models were trained on data from 126–9,327 compounds per target.
  • The study was funded by approximately $X amount of funding from the NIH National Institute of General Medical Sciences (NIGMS) and NIH National Institute of Environmental Health Sciences (NIEHS).

Sources:

  • NewsRx. Studies from Collaborations Pharmaceuticals Inc. Have Provided New Data on Machine Learning (Machine Learning and Large Language Models for Modeling Complex Toxicity Pathways and Predicting Steroidogenesis). Health & Medicine Week. July 25, 2025; p 4474.
  • Environmental Science & Technology. Machine Learning and Large Language Models for Modeling Complex Toxicity Pathways and Predicting Steroidogenesis. 2025.
  • ChEMBL database. A publicly available database of bioactive compounds.
  • Collaborations Pharmaceuticals Inc. For more information on this research, contact Thomas R. Lane at Raleigh, NC 27606, United States.

Read more