Development of homology model, docking protocol and Machine-Learning based scoring functions for identification of Equus caballus’s butyrylcholinesterase inhibitors.

The researchers developed and validated a machine learning model to predict inhibitors of the horse enzyme butyrylcholinesterase (BChE), which has similar properties to human BChE. This model aims to improve the prediction of potential drug toxicity and suitable drug inhibitors for BChE.

Homology Model and Docking Protocol

The researchers started with the development of a homology model that can mimic the actual behavior of horse BChE enzyme. This model is based on the similarity or homology between the horse and human BChE.

• A docking protocol was also developed to forecast how potential drugs can bind to the enzyme.
• The validation of the model was determined by comparing the predicted interactions with the results of a similar experiment done on human BChE. A high degree of similarity indicated that the model has been validated.
• The researchers used a tool called Cavityplus to determine the active site residues of the enzyme, and this information was used to generate a grid box for docking.

Scoring Function

After developing and validating the model, the next step involved creating a scoring function (SF).

• The SF was used to score potential inhibitors based on how well they could bind to the BChE enzyme.
• The researchers tested for the validity of the SF by comparing its scores to known BChE inhibitors.
• The binding energies obtained from SF were correlated with IC values of inhibitors, although the initial results suggested there was poor predictivity from the native SF.

Machine Learning-Based Scoring Functions

Due to the poor predictivity of the native SF, the researchers decided to develop machine learning models to improve the scoring function.

• Information about protein-ligand binding energy, interaction profile, and ligand descriptors were fed into these machine learning algorithms.
• Several models including classification, regression, random forest and extra tree were developed and validated.
• The research team found that their machine learning models gave better results compared to the Autodock scoring function, with an improved area under the curve for the receiver operating characteristic (ROC).
• The final validated model was made available to the public through a web application and a Python library.