Interpretable discovery of patterns in tabular data via spatially semantic topographic maps

Area of Science:

• Computational Biology
• Data Science
• Machine Learning

Background:

• Tabular data is prevalent but challenging for discovering underlying associations.
• Existing representations hinder effective data analysis and pattern discovery.

Purpose of the Study:

• To develop a broadly applicable strategy for unravelling intertwined relationships in tabular data.
• To introduce TabMap, a novel method for reconfiguring data samples into spatially semantic 2D topographic maps.

Main Methods:

• Data samples are reconfigured into 2D topographic maps (TabMap) preserving feature values as pixel intensities.
• Spatial encoding of feature relationships within the map.
• Application of 2D convolutional neural networks for pattern extraction and feature importance ranking.

Main Results:

• TabMap enables the application of 2D convolutional neural networks to tabular data.
• Demonstrated superior predictive performance across 12 diverse biomedical datasets.
• Provided interpretability through feature importance ranking.

Conclusions:

• TabMap is a broadly applicable strategy for uncovering complex relationships in tabular data.
• The method significantly enhances data analysis, pattern discovery, and predictive modeling.
• Successfully applied to disease diagnosis, activity recognition, microbial identification, and QSAR analysis.