Predicting multiple drugs side effects with a general drug-target interaction thermodynamic Markov model

Area of Science:

• Computational Chemistry
• Pharmacology
• Bioinformatics

Background:

• Current molecular descriptors primarily focus on molecular structure, neglecting drug targets or toxicological effects.
• There is a need for advanced computational models that integrate molecular structure with biological system interactions to predict drug-induced side effects.

Purpose of the Study:

• To develop and validate a novel Markov chain-based molecular descriptor that incorporates both molecular structure and affected biological systems.
• To create a comprehensive model capable of predicting a wide range of drug side effects across different organ systems.

Main Methods:

• Utilized Markov chain models to define new molecular descriptors.
• Developed a general Markov model encompassing 39 drug side effects across 11 affected systems for 301 drugs.
• Employed linear discriminant analysis (LDA) with forward stepwise selection for classification and variable selection.

Main Results:

• Achieved high classification accuracy across various organ systems, with rates ranging from 88% to 100% for training/prediction sets.
• Demonstrated the model's ability to predict specific drug side effects, such as a decrease in lymphocytes after administration of antibacterial drug G-1.
• Successfully integrated a large number of side effects into a single stochastic framework.

Conclusions:

• The developed Markov model provides a novel approach to molecular descriptor design, enhancing prediction of drug side effects by considering target systems.
• This framework offers a powerful tool for understanding and predicting drug toxicity and adverse events.
• The study highlights the potential of stochastic models in bridging molecular structure with complex biological outcomes.