Deep convolutional neural networks for annotating gene expression patterns in the mouse brain.

Area of Science:

• Neuroscience
• Computational Biology
• Genomics

Background:

• Understanding gene regulation in brain development requires gene expression profiling across spatial and temporal scales.
• The Allen Developing Mouse Brain Atlas offers high-resolution 3-D in situ hybridization (ISH) data for developing mouse brains.
• Manual annotation of ISH images is time-consuming and limits scalability.

Purpose of the Study:

• To develop and evaluate a computational approach for annotating gene expression patterns in the developing mouse brain.
• To leverage deep learning for automated feature extraction and classification of ISH images.
• To assess the performance of the proposed method against baseline approaches.

Main Methods:

• Utilized deep convolutional neural networks (CNNs) trained on natural images for feature extraction from ISH images.
• Employed invariant image feature descriptors and bag-of-words models as baseline representations.
• Combined features from multiple ISH sections to create 3-D, brain-wide gene expression representations.
• Applied regularized learning methods for discriminating gene expression patterns across brain structures.

Main Results:

• The CNN-based approach significantly outperformed baseline methods in annotating gene expression patterns.
• Achieved an average Area Under the Curve (AUC) of 0.894 ± 0.014, compared to 0.820 ± 0.046 for the bag-of-words approach.
• Demonstrated superior performance across multiple brain structure levels and four developmental ages.

Conclusions:

• Deep convolutional neural networks show strong transfer learning capabilities for biological image analysis.
• The proposed computational method provides an effective and accurate solution for annotating developing mouse brain gene expression patterns.
• This automated approach facilitates large-scale analysis in developmental neuroscience.