Convolutional Embedded Networks for Population Scale Clustering and Bio-Ancestry Inferencing.

Area of Science:

• Genomics
• Bioinformatics
• Machine Learning

Background:

• Genetic variants (GVs) are crucial for understanding population structure, disease susceptibility, and drug response.
• Machine learning, particularly deep neural networks (DNNs), shows promise in identifying complex interactions among GVs.
• Effective data representation is key for high-performance machine learning algorithms in genomics.

Purpose of the Study:

• To develop and evaluate a novel deep learning framework, convolutional embedded networks (CEN), for analyzing large-scale genetic variant data.
• To enhance the clustering of individuals into population groups and the prediction of geographic ethnicity based on GVs.
• To offer a transparent, scalable, and efficient alternative to existing methods for genomic data analysis.

Main Methods:

• Proposed convolutional embedded networks (CEN), integrating convolutional embedded clustering (CEC) for population clustering and convolutional autoencoder (CAE) for ethnicity prediction.
• Applied CAE-based representation learning to 95 million GVs from the '1000 genomes' and 'Simons genome diversity' projects.
• Utilized gradient boosted trees (GBT) and SHapley Additive exPlanations (SHAP) for identifying significant biomarkers and interpreting model predictions.

Main Results:

• CEC achieved high clustering performance with an Adjusted Rand Index (ARI) of 0.915, Normalized Mutual Information (NMI) of 0.92, and Clustering Accuracy (ACC) of 89% within 22 hours.
• The CAE classifier demonstrated strong predictive power for geographic ethnicity, yielding an F1 score of 0.9004 and a Mathews Correlation Coefficient (MCC) of 0.8245.
• The proposed CEN approach outperformed state-of-the-art methods like VariantSpark and ADMIXTURE in accuracy and scalability.

Conclusions:

• Convolutional embedded networks (CEN) provide an accurate, efficient, and scalable solution for analyzing genetic variants.
• The method enables precise population clustering and geographic ethnicity prediction, aiding in understanding human genetic diversity.
• The interpretability features enhance the utility of GVs analysis for biomarker discovery and personalized medicine.