Assessing AlphaFold AI’s Protease Enzyme Structure Prediction Accuracy

Abstract

This study analyzed AlphaFold AI’s ability in accurately predicting protease enzyme structures. AlphaFold uses machine learning, taking amino acid sequences and using physical and scientific knowledge of protein structures to generate a protein structure prediction. Past studies have comfirmed AlphaFold’s general abilities but have identified limitations in certain factors, like post translational modifications, ligands, and other environmental factors. However, there have not been studies assessing AlphaFold in predicting protease enzyme structures specifically. Quantitative data was collected using ex-post facto and correlational methods, which compared the RMSD score between AlphaFold and Protein Data Bank structures of the same protease enzyme. Furthermore, correlational trends were searched for between protein complexity and length with the RMSD score. 77% of the 30 protease enzymes assessed were found to be accurate, with more complex structures lowering in accuracy. Protein length was not a factor in AlphaFold’s prediction accuracy. By utilizing these findings, researchers in the pharmaceutical industry can consider the weak points of AlphaFold, conduct further studies identifying more factors that contribute to AlphaFold’s accuracy, and work on improving the program based on the results.