Investigation of Machine Intelligence in Compound Cell Activity Classification.

Area of Science:

• Computational chemistry
• Machine learning in drug discovery

Background:

• Machine intelligence is increasingly vital in drug discovery.
• Predicting compound cell activity is crucial for identifying potential therapeutics.

Purpose of the Study:

• To explore machine intelligence methods for cell activity prediction.
• To evaluate different machine learning and deep learning models using simplified molecular input line entry system strings as direct input.

Main Methods:

• Employed multiple machine intelligence algorithms: support vector machine, decision tree, random forest, extra trees, gradient boosting machine, convolutional neural network, long short-term memory network, and gated recurrent unit network.
• Utilized simplified molecular input line entry system strings as direct input, mimicking natural language processing.
• Evaluated models on single and whole datasets with balanced and imbalanced distributions, using nine performance metrics.

Main Results:

• Gradient boosting machine demonstrated competence on balanced datasets.
• Convolutional neural network proved effective for imbalanced datasets.
• Both classic machine learning and deep learning methods showed potential for compound cell activity classification.

Conclusions:

• Machine intelligence, particularly gradient boosting and convolutional neural networks, offers a powerful approach to cell activity prediction.
• Directly using simplified molecular input line entry system strings simplifies the process and shows promising results.
• This study highlights the potential of AI in accelerating drug discovery pipelines.