AcrNET

AcrNET predicts anti-CRISPR (Acr) proteins and their classes using deep neural networks and transformer-derived protein language features to identify inhibitors of CRISPR-Cas systems.

Key Features:

• Deep Neural Network Architecture: Employs a novel deep neural network tailored for anti-CRISPR analysis and achieves at least a 15% improvement in F1 score in cross-dataset validation tests.
• Class Prediction Capability: Predicts detailed classes of anti-CRISPR proteins to provide insights into mechanisms of CRISPR-Cas inhibition.
• Integration with Transformer Models: Leverages the ESM-1b protein language model pre-trained on 250 million protein sequences to capture complex sequence patterns.
• Complementary Feature Analysis: Combines transformer-derived features, evolutionary characteristics, and local structural properties that work synergistically to enhance prediction robustness.
• Validation via Structure and Interaction Analyses: Uses AlphaFold for protein structure prediction and conducts motif analysis and docking experiments to validate conserved patterns and implicit interactions.
• Data-Scarcity Mitigation: Addresses data scarcity by incorporating pre-trained transformer features to improve generalization on limited labeled anti-CRISPR data.

Scientific Applications:

• Gene editing: Supports identification and classification of Acrs to inform control and modulation strategies for CRISPR-based gene editing.
• Phage therapy: Informs development of phage therapy approaches by predicting Acrs that modulate bacterial CRISPR immunity.
• Mechanistic elucidation of CRISPR-Cas inhibition: Aids discovery of conserved motifs, structural features, and interaction patterns that clarify mechanisms of inhibition.

Methodology:

Integrates a novel deep neural network with transformer-derived features from ESM-1b (pre-trained on 250 million sequences), evolutionary features, and local structural information; validated via cross-dataset F1 evaluation, AlphaFold structure prediction, motif analysis, and docking experiments.