A steroids QSAR approach based on approximate similarity measurements

Area of Science:

• Computational Chemistry
• Medicinal Chemistry
• Drug Discovery

Background:

• Quantitative Structure-Activity Relationship (QSAR) models are crucial for predicting drug efficacy.
• Traditional similarity metrics in QSAR may limit predictive accuracy.
• Developing novel similarity measures can enhance QSAR model performance.

Purpose of the Study:

• To introduce a new QSAR methodology based on approximate similarity measurements.
• To enhance the predictive power of QSAR models by incorporating nonisomorphic subgraph distances.
• To apply the developed method for predicting steroid binding to the corticosteroid globulin receptor.

Main Methods:

• Calculating approximate similarity using graph isomorphism and nonisomorphic subgraph distances.
• Employing a parametric function with topological invariants for distance computation.
• Optimizing the contribution of nonisomorphic subgraph distances to overall similarity.
• Building predictive models using approximate similarity matrices and validating through cross-validation.

Main Results:

• Predictive models built with approximate similarity matrices exhibit higher predictive ability compared to traditional methods.
• The method achieved strong external prediction results for steroid binding (r=0.82, SEP=0.30).
• Cross-validation yielded excellent performance metrics (Q2=0.84, SECV=0.47).

Conclusions:

• The novel QSAR method based on approximate similarity offers improved predictive performance.
• This approach provides a valuable tool for predicting molecular interactions, such as steroid-receptor binding.
• The optimized method enhances the accuracy and reliability of QSAR-driven drug discovery efforts.