{
  "id": 3761719,
  "title": "PMPNN-DDG: an accurate machine learning-based {triangleup}{triangleup}G prediction pipeline trained on a novel interpretable feature set extracted from ProteinMPNN",
  "url": "https://urgent.news/2026/08/27/pmpnn-ddg-an-accurate-machine-learning-based-triangleup-triangleup-g",
  "topic": "ai",
  "section": "AI",
  "published": "2026-08-27T00:00:00.000Z",
  "source": {
    "name": "bioRxiv",
    "slug": "biorxiv",
    "url": "https://www.biorxiv.org/content/10.64898/2026.08.23.746499v1?rss=1"
  },
  "original_language": "en",
  "account": "Researchers have developed a new method called PMPNN-DDG for accurately predicting the impact of single-point mutations on protein thermodynamic stability, known as DDG. This is important for understanding how genetic variations can lead to changes in protein function. Previous attempts to solve this problem have faced challenges due to limited training data and difficulty in predicting the magnitude of structural changes. To improve upon these methods, the researchers created PMPNN-DDG, a Random Forest-based predictor that utilizes unique and interpretable features derived from a model called ProteinMPNN. By training PMPNN-DDG on this novel data set, the researchers found it outperformed existing methods in several evaluation metrics. Specifically, on one test set, PMPNN-DDG achieved an rF +R value of 0.64 and an RMSE of 1.45, surpassing other models. On a different test set, it performed even better with an rF +R of 0.81, an rF -R of -0.99, and an RMSE of 1.10. These results demonstrate that PMPNN-DDG offers a more accurate and reliable approach to predicting the effects of protein mutations, and the model is now freely accessible online for others to use and build upon.",
  "summary": "An accurate and tractable approximation of the single-point mutation-induced change in protein thermodynamic stability, denoted by DDG, is critical for understanding the genotype-phenotype relationship. Several computational methods have been proposed for this problem; however, limited and error-prone training data and the difficult-to-predict magnitude of structural perturbations make this a…",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}