Development of a machine learning-based target-specific scoring function for structure-based binding affinity...

Area of Science:

• Biochemistry and enzymology
• Computational chemistry and drug discovery
• Medicinal chemistry

Background:

• Human dihydroorotate dehydrogenase (hDHODH) is crucial for de novo pyrimidine synthesis.
• hDHODH is a validated therapeutic target for autoimmune disorders and cancers.
• Existing scoring functions have limitations in accurately predicting hDHODH inhibitor efficacy.

Purpose of the Study:

• To develop a novel, target-specific scoring function (TSSF-hDHODH) for human dihydroorotate dehydrogenase.
• To enhance the accuracy of virtual screening for hDHODH inhibitors.
• To identify potential drug candidates for hDHODH-related diseases.

Main Methods:

• Utilized docking structures from AutoDock Vina.
• Integrated enzyme-ligand interaction and ligand features.
• Employed support vector regression to build the TSSF-hDHODH scoring function.
• Validated the scoring function using cross-validation and external datasets.
• Performed virtual screening on the FDA-Approved & Pharmacopeia Drug Library.
• Conducted molecular dynamics simulations for candidate validation.

Main Results:

• The TSSF-hDHODH scoring function achieved high Pearson correlation coefficients (0.86 in cross-validation, 0.74 in external validation).
• TSSF-hDHODH significantly outperformed AutoDock Vina and RF-Score.
• Virtual screening identified crizotinib as a potential hDHODH inhibitor.
• Molecular dynamics simulations supported crizotinib's candidacy for further optimization.

Conclusions:

• The developed TSSF-hDHODH scoring function is a powerful tool for identifying hDHODH inhibitors.
• This approach facilitates drug discovery for autoimmune diseases and cancer.
• The methodology can be extended to develop scoring functions for other enzyme targets.