MLCNNF: A Multi-Label Convolutional Neural Network Framework for Predicting Adverse COVID Drug Reactions From the...

Area of Science:

• Drug Discovery and Development
• Computational Chemistry
• Artificial Intelligence in Medicine

Background:

• Limited effective treatments for COVID-19 necessitate careful drug development.
• Adverse COVID Drug Reactions (ACDR) pose significant risks, often undetected during clinical trials.
• Existing methods for predicting adverse drug reactions rely on surveys, biological factors, or textual data, leaving chemical structure analysis unexplored.

Purpose of the Study:

• To investigate the potential of using 2D chemical drug structures for predicting adverse COVID drug reactions (ACDR).
• To develop and validate a novel computational framework for multi-label prediction of ACDR.
• To assess the efficacy of the proposed method against established transfer learning models.

Main Methods:

• A Multi-Label Convolutional Neural Network Framework (MLCNNF) was developed, utilizing image augmentation.
• Chemical 2D drug structures were represented using RGB color channels.
• Convolution2D and MaxPooling2D layers were employed for feature extraction, with validation via leave-one-out cross-validation.

Main Results:

• The MLCNNF model achieved a high accuracy rate of 99.40 ± 0.0344%.
• The proposed framework demonstrated superior performance compared to multiple transfer learning models, including DenseNet201, VGG19, ResNet50, and others.
• Prediction of ACDR directly from chemical 2D structures proved effective.

Conclusions:

• 2D chemical drug structures contain predictive information for adverse COVID drug reactions (ACDR).
• The MLCNNF framework offers a promising and accurate approach for identifying potential ACDR early in drug development.
• This method can aid in the selection of safer drug candidates for COVID-19 treatment.