Assessing AlphaFold 3 for Per- and Polyfluoroalkyl Substances Docking in Protein Structures.

Area of Science:

• Environmental Chemistry
• Computational Biology
• Toxicology

Background:

• Per- and polyfluoroalkyl substances (PFAS) are persistent environmental pollutants with potential health risks linked to protein interactions.
• Accurate modeling of protein-PFAS interactions is crucial for understanding their toxicological mechanisms.
• AlphaFold 3 (AF3) is a novel tool for protein-ligand modeling, but its efficacy with PFAS requires evaluation.

Purpose of the Study:

• To assess the reliability of AlphaFold 3 (AF3) in predicting protein-PFAS interactions.
• To evaluate AF3's performance on both training-seen and unseen PFAS-protein datasets.
• To investigate the influence of PFAS chemical structure on docking outcomes.

Main Methods:

• Utilized a curated Protein Data Bank dataset of protein-PFAS interactions, split into "Before" (training) and "After" (unseen) sets.
• Evaluated AF3's accuracy in predicting protein structures, binding pockets, and ligand poses.
• Assessed the impact of PFAS chemical properties on predicted binding modes.
• Compared AF3 performance with a hybrid approach combining AF3 and Vina.

Main Results:

• AF3 accurately predicted protein structures and binding pockets.
• Ligand pose prediction accuracy decreased significantly for the unseen "After Set" (55.8%) compared to the "Before Set" (74.5%), suggesting potential overfitting.
• AF3 favored headgroup interactions with polar/charged residues, differing from some native binding modes.
• A hybrid AF3-Vina approach, considering multiple poses, enhanced prediction accuracy.

Conclusions:

• AF3 demonstrates high accuracy for protein structure and pocket prediction but requires caution for PFAS ligand docking, especially for novel compounds.
• The study highlights potential overfitting in AF3 for PFAS interactions.
• Combining AF3 with Vina offers a more robust strategy for accurate protein-PFAS interaction modeling.