What question did this study set out to answer?

The research aims to improve the drug discovery process for breast cancer by utilizing AI frameworks for solubility prediction and lead selection.

April 22, 2026Open Access

Title: Learning, Simulating, and Inhibiting: A Multi-Scale AI Framework for Breast Cancer Drug Discovery

Puntos clave

The research aims to improve the drug discovery process for breast cancer by utilizing AI frameworks for solubility prediction and lead selection.
Implemented machine learning models (RF, XGBoost, LightGBM, ANN) for solubility prediction with over 80% accuracy.
Conducted virtual screening against estrogen receptor alpha ligand binding domain Y537S with selected hits.
Performed molecular dynamics simulations and computed binding free energies using MMGBSA and MMPBSA.
Top hits showed favorable binding free energy: Hit-1 at -111.34 kcal/mol and Hit-3 at -112.1 kcal/mol.
ADMET predictions indicated potential for the selected compounds as promising leads.
The protein-ligand complexes demonstrated stability and favorable dynamics during molecular simulations.

Resumen

Cancer is one of the most prevalent malignant diseases in the world, as breast cancer (BC) ranks as the second most frequent cause of death in women. Solubility is of great importance in drug research and development. Ensuring that molecules with the highest levels of solubility are prioritized in the initial phases of drug discovery is important in reducing the amount of resources and increasing the chances of clinical success. Herein, RF, XGBoost, LightGBM, and Artificial Neural Networks machine learning and deep learning models demonstrated over 80 percent accuracy in solubility prediction during training and testing phases. Virtual screening using a structure-based method was used against estrogen receptor alpha ligand binding domain Y537S breast cancer (BC). Best-ranked leads were selected: Hit-1 (-10 kcal/mol), Hit-2 (-9.4 kcal/mol), Hit-3 (-9.2 kcal/mol), in contrast to the control (-6 kcal/mol). ADMET (Absorption, Distribution, Metabolism, Excretion, and Toxicity) predictions were done for the top selected leads. Additionally, the binding mechanisms were assessed through molecular dynamics simulations over 150 ns. MMGBSA binding free energy of complexes was in the following order: Hit-1 (-111.34 kcal/mol), Hit-2 (-105.02 kcal/mol), Hit-3 (-112.1 kcal/mol), and control (-99.71 kcal/mol). While MMPBSA results were -107.97 kcal/mol, -103.48 kcal/mol, -107.01 kcal/mol and -97.78 kcal/mol, respectively. A dynamic cross correlation (DCCM), radial distribution function (RDF), principal component analysis (PCA), and free energy landscape (FEL) were computed, and clusters for the hit complexes were analyzed. Salt bridge interactions and secondary structure analyses indicated the identified compounds as promising binding leads. The protein-ligand complexes exhibited stable and favorable dynamics and thus can be subjected to further research to explore their breast cancer inhibition potential.Keywords: Deep learning, ML, Breast Cancer, Molecular docking, MD simulation, MMPBSA/GBSA.

Leer artículo completoexternamente

Preguntar a la IA

Me gusta

Guardar

Ver artículo completo

Cite This Study

M Aljohani Hashim (Mon,) studied this question.

synapsesocial.com/papers/69e866c96e0dea528ddeb2e0 https://doi.org/https://doi.org/10.5281/zenodo.19657593

Preguntar a la IA

Me gusta

Guardar

Ver artículo completo