DeepInsight: A methodology to transform a non-image data to an image for convolution neural network architecture.

Area of Science:

• Bioinformatics
• Machine Learning
• Data Science

Background:

• Extracting subtle variations from large, complex datasets (genomic, text, etc.) is challenging.
• Scattered information across genes or data elements hinders identification of underlying mechanisms.
• Clustering similar data elements can improve accessibility and reveal hidden patterns.

Purpose of the Study:

• To introduce DeepInsight, a novel method for converting non-image data into an image format.
• To leverage Convolutional Neural Networks (CNNs) for analyzing non-image datasets.
• To enable robust feature extraction and identification of critical variations in diverse data types.

Main Methods:

• DeepInsight converts diverse non-image samples (RNA-seq, text, artificial data) into image representations.
• Convolutional Neural Networks (CNNs) are applied to these image-formated datasets.
• The method utilizes GPU acceleration for efficient computation.

Main Results:

• DeepInsight successfully enables CNN application to non-image data.
• The method demonstrates promising results in feature extraction and identification of imperative information.
• This represents the first application of CNNs across multiple non-image datasets simultaneously.

Conclusions:

• DeepInsight offers a powerful new approach for analyzing complex non-image data.
• The method enhances the utility of CNNs beyond traditional image analysis.
• This technique facilitates the discovery of subtle variations and hidden mechanisms in various data domains.