Expectation pooling

Expectation pooling integrates probabilistic expectation-based global pooling with convolutional neural networks (CNNs) to improve the accuracy and interpretability of DNA–protein binding predictions.

Key Features:

• Statistical foundation: Derived from the expectation–maximization algorithm, providing a probabilistic basis for the pooling operation.
• Integration with CNNs: Combines probabilistic pooling with convolutional neural networks (CNNs) to leverage CNNs' sequence-specific modeling capabilities.
• Global pooling: Implements a global pooling approach that aggregates convolutional outputs via expectation calculations.
• Interpretability: Computes expectations of inputs to preserve interpretability without increasing model parameters.
• Parameter efficiency: Enhances prediction performance without adding model complexity or extra parameters.
• Hyperparameter analysis: Includes analysis of hyperparameters and offers optional pooling structures tailored to different datasets.

Scientific Applications:

• DNA–protein binding prediction: Improves prediction of sequence specificity for DNA–protein interactions.
• Binding site identification and regulatory mechanism analysis: Aids identification of potential binding sites and understanding of gene regulatory mechanisms within genomic sequences.

Methodology:

Expectation pooling applies expectation calculations as a global pooling layer within CNN architectures, is derived from expectation–maximization principles, computes expectations of inputs without increasing model parameters, and is validated by experimental comparisons for DNA–protein binding prediction.