miTAR

miTAR predicts microRNA (miRNA) target genes from sequence using a hybrid deep learning approach that integrates convolutional neural networks and recurrent neural networks to capture spatial and sequential features of miRNA:target interactions.

Key Features:

• Hybrid deep learning architecture: Integrates convolutional neural networks (CNNs) and recurrent neural networks (RNNs) to model spatial and sequential aspects of miRNA:target interactions.
• Input data: Uses raw sequences of miRNAs and gene targets as model inputs.
• Training datasets: Trained on the DeepMirTar and miRAW datasets.
• Model variants: Trains separate models for each dataset and a unified model combining data from both datasets.
• Evaluation metrics: Performance evaluated using accuracy, sensitivity, specificity, F-score, and Brier Score.
• Reported performance: Achieves accuracy 0.9787 versus 0.9348 for DeepMirTar, 0.9649 versus 0.935 for miRAW, and a unified model accuracy of 0.9545.
• Comparative performance: Demonstrates higher accuracy and improved metrics compared to earlier reported methods on the same datasets.

Scientific Applications:

• miRNA target prediction: Predicts candidate target genes for miRNAs from sequence data.
• Investigation of miRNA-mediated regulation: Facilitates studies of post-transcriptional and translational regulation by miRNAs.
• Method benchmarking: Provides comparative performance data for evaluating deep learning approaches in miRNA target prediction research.

Methodology:

miTAR uses raw miRNA and gene target sequences as inputs, integrates CNNs and RNNs in a hybrid model, trains separate models on DeepMirTar and miRAW and a unified model combining both datasets, and evaluates performance with accuracy, sensitivity, specificity, F-score, and Brier Score.