Introducing a foundational sequence transformer for range adaptive nucleotide decoding (STRAND).

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• High-throughput sequencing generates vast genomic data, necessitating advanced analytical models.
• Accurate variant detection and interpretation are crucial for understanding human health and disease.

Purpose of the Study:

• To introduce a novel exomic foundational model for enhanced variant detection and interpretation.
• To evaluate the model's performance against existing benchmarks in downstream genomic tasks.

Main Methods:

• Developed a short-range transformer architecture trained on human exomic sequences from the Tapestry study.
• Incorporated multispecies data alongside the human reference genome.
• Utilized a curated exomic ClinVar dataset for pathogenicity and disease state evaluation.

Main Results:

• The model demonstrated high accuracy in predicting next token accuracy and identifying clinically pathogenic variants.
• The largest model (1B parameters) achieved a mean accuracy of 0.880, outperforming previous benchmarks.
• The model showed superior performance in variant effect prediction and disease state identification.

Conclusions:

• The novel exomic foundational model significantly improves variant detection and interpretation.
• This advancement has major implications for genomics-based diagnosis and personalized medicine.
• The model facilitates tailored therapeutic development and a deeper understanding of the human exome.