Omics-CNN: A comprehensive pipeline for predictive analytics in quantitative omics using one-dimensional...

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning in Healthcare

Background:

• Machine learning models are crucial for predicting severe diseases.
• Convolutional Neural Networks (CNNs) are sophisticated tools with potential for omics data classification.
• Applying CNNs to multidimensional omics data requires specialized approaches.

Purpose of the Study:

• To expand Convolutional Neural Networks (CNNs) for multidimensional omics data classification.
• To develop Omics-CNN, a pipeline for accurate and interpretable classification models from high-throughput omics data.
• To identify diagnostic biomarkers for diseases like Ischemic Stroke and COVID-19.

Main Methods:

• Introduced Omics-CNN, a pipeline coupling CNNs with dimensionality reduction, preprocessing, clustering, and explainability techniques.
• Compared univariate and multivariate dimensionality reduction techniques.
• Utilized Gradient Weighted Class Activation Mapping for feature importance analysis.

Main Results:

• Applied Omics-CNN to transcriptomics and proteomics data for Ischemic Stroke (IS) and COVID-19.
• Achieved high accuracies for diagnostic models: 96% for IS and 95.41% for COVID-19.
• Identified key biosignatures for IS (KRT15, VPRBP, TNFRSF4, GORASP2) and COVID-19 (ADGRB3, VNN2, AGER, CIAPIN1).

Conclusions:

• Omics-CNN effectively addresses challenges in applying CNNs to quantitative omics data.
• The pipeline outperforms previous machine learning models for IS and COVID-19 diagnosis.
• Omics-CNN successfully determines the most contributing biomarkers for disease diagnosis.