Multiple machine learning based descriptive and predictive workflow for the identification of potential PTP1B...

Area of Science:

• Biochemistry
• Computational Chemistry
• Pharmacology

Background:

• Protein-Tyrosine Phosphatase 1B (PTP1B) is a critical negative regulator in insulin and leptin signaling pathways.
• PTP1B is a significant therapeutic target for managing Type-2 Diabetes (T2D) and obesity.

Purpose of the Study:

• To develop and optimize machine learning classification models for identifying novel PTP1B inhibitors.
• To utilize predictive modeling for virtual screening of compound libraries and experimental validation.

Main Methods:

• Employed machine learning techniques including Naïve Bayesian, Random Forest, Support Vector Machine, and K-Nearest Neighbors.
• Utilized structural fingerprints and molecular descriptors for model construction and optimization.
• Performed virtual screening on the Maybridge small compound database using the best predictive model.

Main Results:

• Multiple models achieved over 90% prediction accuracy on training and test sets.
• The best Support Vector Machine model demonstrated a Matthews Correlation Coefficient of 0.82 on an external test set.
• Experimental assays confirmed PTP1B inhibitory activity in two out of five screened compounds, with identified key structural fragments.

Conclusions:

• The developed machine learning strategy effectively identifies PTP1B inhibitors from large compound libraries.
• This approach aids in the design of new therapeutic molecules targeting PTP1B for T2D and obesity.
• The study highlights the potential of computational methods in drug discovery for metabolic diseases.