Machine Learning of Three-Dimensional Protein Structures to Predict the Functional Impacts of Genome Variation.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Human genome research generates vast genetic data, increasing variants of unknown significance (VUS).
• Current bioinformatics tools effectively identify rare variants and disease associations but lack robust biological activity prediction for VUS.
• Clinical genetic interpretation and patient treatment stratification are impacted by the challenge of VUS.

Purpose of the Study:

• To develop a computational method for predicting the biological impact of genetic variants.
• To address the gap in predicting variant effects on gene function using protein structure information.
• To identify key protein regions influencing transcriptional activities of proto-oncogenes NRF2 and c-Myc.

Main Methods:

• Developed a machine learning method integrating AlphaFold-predicted protein 3D structures.
• Trained state-of-the-art machine learning classifiers to predict variant impact on transcriptional activities.
• Focused on two key proto-oncogenes: nuclear factor erythroid 2 (NFE2L2)-related factor 2 (NRF2) and c-Myc.

Main Results:

• Achieved prediction accuracies exceeding 80% for variant impact on transcriptional pathways.
• Identified specific protein regions critical for perturbing NRF2 and c-Myc pathway activities.
• Demonstrated the utility of protein structure-based machine learning for variant effect prediction.

Conclusions:

• The developed machine learning method effectively predicts the functional impact of genetic variants using protein structural data.
• This approach enhances the interpretation of variants of unknown significance (VUS) in clinical genetics.
• The findings provide insights into key regulatory regions of NRF2 and c-Myc, aiding in understanding disease mechanisms.