Leveraging transformers for semi-supervised pathogenicity prediction with soft labels.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Next-Generation Sequencing (NGS) generates vast genomic data requiring advanced analytical methods.
• Accurate prediction of genetic variant pathogenicity is crucial for personalized medicine.

Purpose of the Study:

• To develop and evaluate a Deep Learning model for predicting genetic variant pathogenicity.
• To leverage semi-supervised learning for efficient utilization of diverse genetic variant data.

Main Methods:

• A Feature Tokenizer Transformer architecture was employed to process numerical and categorical genomic data.
• A semi-supervised learning approach was utilized on NGS-derived datasets.
• Data preprocessing included imputation, scaling, and encoding for quality assurance.

Main Results:

• The Deep Learning model demonstrated high accuracy in predicting the pathogenicity of confidently labeled genetic variants.
• The study assessed the model's performance on less certain (soft-labeled) genetic variants.
• The Feature Tokenizer Transformer effectively handled heterogeneous genomic data types.

Conclusions:

• The developed Deep Learning model shows significant promise for accurate genetic variant pathogenicity prediction.
• This approach can enhance the interpretation of NGS data for clinical applications.
• Semi-supervised learning and advanced architectures improve genomic data analysis for personalized medicine.