An Inverse QSAR Method Based on Linear Regression and Integer Programming.

Area of Science:

• Computational chemistry
• Medicinal chemistry
• Drug discovery

Background:

• Computer-aided drug design utilizes inverse quantitative structure-activity relationship (QSAR) to infer chemical compounds from activity data.
• Existing inverse QSAR methods often fail to guarantee exact or optimal solutions.

Purpose of the Study:

• To develop a novel computational framework for designing chemical structures.
• To improve upon existing methods for inferring chemical compounds based on activity and constraints.

Main Methods:

• A novel framework combining linear regression and mixed integer linear programming (MILP) was developed.
• Linear regression was used to construct a prediction function, replacing artificial neural networks (ANNs).
• A novel MILP formulation was derived to simulate the linear regression prediction function's computation.

Main Results:

• The proposed method demonstrated good prediction accuracy for 18 chemical properties, outperforming ANNs in previous work.
• The method successfully inferred chemical structures with up to 50 non-hydrogen atoms for five chemical properties.

Conclusions:

• The combination of linear regression and integer programming presents a promising approach for computational molecular design.
• This method offers a viable alternative for designing novel chemical structures with desired properties.