Efficient HLA imputation from sequential SNPs data by transformer.

Area of Science:

• Genetics and Genomics
• Bioinformatics
• Immunogenetics

Background:

• Human leukocyte antigen (HLA) gene typing is crucial for disease association studies but is costly and time-consuming.
• Existing imputation methods using single nucleotide polymorphisms (SNPs) data, including deep learning models like DEEP*HLA, struggle with infrequent alleles and large datasets.

Purpose of the Study:

• To develop a novel Transformer-based model, HLARIMNT, for accurate and efficient HLA allele imputation.
• To evaluate HLARIMNT's performance against existing methods, particularly for rare alleles and varying training data sizes.

Main Methods:

• Developed HLARIMNT, a Transformer-based model leveraging the sequential nature of SNPs data for HLA imputation.
• Evaluated HLARIMNT using Pan-Asian and Type 1 Diabetes Genetics Consortium (T1DGC) reference panels.
• Compared HLARIMNT's imputation accuracy with the DEEP*HLA model across different training data volumes.

Main Results:

• HLARIMNT demonstrated superior imputation accuracy compared to DEEP*HLA, notably for infrequent HLA alleles.
• The model's performance remained robust across different reference panel sizes and training data volumes.
• Transformer architecture effectively utilizes sequential SNPs data for enhanced imputation.

Conclusions:

• HLARIMNT offers a significant advancement in HLA allele imputation, providing higher accuracy and efficiency.
• Transformer-based models show promise for imputing not only HLA types but potentially other gene types from sequential SNPs data.
• This approach could reduce the cost and time associated with genetic typing in large-scale studies.