Weka machine learning for predicting the phospholipidosis inducing potential.

Area of Science:

• Computational chemistry
• Drug discovery
• Toxicology

Background:

• Drug discovery is a lengthy and costly process, often exceeding 18 years and $2 billion USD.
• Computer-assisted drug design (CADD) can significantly reduce drug discovery timelines and expenses.
• Phospholipidosis, an adverse drug effect, involves intracellular phospholipid accumulation and lamellar body formation, particularly with cationic amphiphilic drugs.

Purpose of the Study:

• To develop and evaluate structure-activity relationship (SAR) models for predicting drug-induced phospholipidosis potential.
• To identify machine learning algorithms capable of accurately discriminating between phospholipidosis-inducing and non-inducing chemicals.

Main Methods:

• Utilized Weka machine learning software to build SAR models.
• Employed various algorithms including k-nearest neighbors, decision trees, support vector machines, and artificial immune systems.
• Trained models to identify drugs with phospholipidosis-inducing potential based on chemical structures.

Main Results:

• Support vector machines (SVMs) provided the most accurate predictions for phospholipidosis-inducing potential.
• Other high-performing algorithms included perceptron artificial neural networks, logistic regression, and k-nearest neighbors.
• The developed SAR models effectively discriminated between chemicals with and without phospholipidosis-inducing properties.

Conclusions:

• Machine learning-based SAR models are effective tools for predicting drug-induced phospholipidosis.
• SVMs offer a promising approach for early identification of potential drug candidates with reduced risk of phospholipidosis.
• Computational methods like SAR can accelerate drug discovery by identifying safer drug candidates earlier in the development process.