AI-enabled genomic pattern discovery in Epstein-Barr virus on Amazon EC2

This article describes a collaboration between a national genomics research institute and Temus using AWS to apply AI models and machine learning to analyze Epstein-Barr virus (EBV) genomes at scale for Nasopharyngeal Carcinoma research.

• Built scalable POC pipeline using Amazon S3 for storage and GPU-powered Amazon EC2 instances with NVIDIA H200 and L40 GPUs
• Preprocessed over 1,300 EBV genome samples from FASTA and FASTQ formats using specialized bioinformatics tools
• Selected Evo2 foundation model with StripedHyena 2 architecture to convert genomic data into numerical embeddings
• Applied PCA for dimensionality reduction and HDBSCAN clustering to identify meaningful patterns without presetting cluster numbers
• Achieved 69–97% cluster purity for disease-type labels with strong statistical significance and cluster stability
• Identified distinct clusters correlating with clinical outcomes including NPC, healthy controls, and Burkitt lymphoma

The project demonstrates how cloud infrastructure combined with AI engineering can accelerate genomic pattern discovery, with next steps focused on validation across larger independent cohorts.