Estimation of Acid Dissociation Constants Using Graph Kernels.

Area of Science:

• Computational chemistry
• Drug discovery
• Chemical informatics

Background:

• The biopharmaceutical profile of a compound is influenced by the dissociation constants (pKa) of its ionizable groups.
• Accurate estimation of pKa values is crucial for predicting drug absorption, distribution, metabolism, and excretion (ADME) properties.
• Existing methods often require computationally intensive structure optimization or rely on complex theoretical frameworks.

Purpose of the Study:

• To develop a novel computational method for estimating acidic and basic dissociation constants.
• To utilize kernel ridge regression and graph kernels for pKa prediction directly from annotated chemical structures.
• To compare the performance of the proposed method against established semi-empirical models.

Main Methods:

• Employed kernel ridge regression, a machine learning technique, for quantitative structure-property relationship (QSPR) modeling.
• Utilized graph kernels to represent chemical structures, capturing topological and chemical information.
• Input data consisted of annotated molecular structure graphs, avoiding the need for 3D conformational analysis or geometry optimization.

Main Results:

• The developed approach demonstrated performance comparable to a semi-empirical model based on frontier electron theory.
• The method successfully estimated dissociation constants using only the annotated structure graph.
• No computationally expensive structure optimization was required, streamlining the prediction process.

Conclusions:

• Kernel ridge regression combined with graph kernels offers an efficient and effective alternative for estimating compound dissociation constants.
• This method simplifies the prediction of key physicochemical properties relevant to biopharmaceutical profiling.
• The approach presents advantages in terms of computational efficiency and ease of implementation compared to traditional methods.