riboflow

riboflow classifies putative riboswitch sequences into 32 ligand classes using deep learning to enable ligand-specific identification for genome annotation, synthetic biology, and antibiotic development.

Key Features:

• High Accuracy Classification: Utilizes convolutional neural networks and bidirectional LSTM recurrent neural networks to classify riboswitch sequences into 32 ligand classes, achieving >99% accuracy and a macro-averaged F-score of 0.96 and outperforming conventional machine learning classifiers.
• Advanced Deep Learning Frameworks: Employs CNN and bidirectional RNN with Long Short-Term Memory to capture complex sequence patterns, with models trained using a stratified train-validate-test approach.
• Dynamic Update Functionality: Integrates newly discovered riboswitch sequences and extends predictive modeling to accommodate additional ligand classes.

Scientific Applications:

• Genetic circuit design: Provides ligand-specificity identification of riboswitches to inform the design and assembly of genetic circuits.
• Antibiotic development: Enables detailed characterization of microbial riboswitches to support next-generation antibiotic discovery.
• Genome annotation and biotechnology: Assists genome annotation and biotechnology workflows by classifying riboswitches and informing synthetic biology applications.

Methodology:

Convolutional neural networks and bidirectional LSTM recurrent neural networks were trained using a stratified train-validate-test split, and the system includes a dynamic update mechanism to integrate new riboswitch sequences.