Related Experiment Video
Updated: Jun 9, 2025

Understanding the Impact of Temperate Bacteriophages on Their Lysogens Through Transcriptomics
Published on: January 5, 2024
A long-context language model for deciphering and generating bacteriophage genomes
Bin Shao1,2, Jiawei Yan3
1Advanced Research Institute of Multidisciplinary Science, Beijing Institute of Technology, Beijing, 100081, China. shaobinlx@gmail.com.
We created megaDNA, a long-context generative model for genomes inspired by large language models (LLMs). This model predicts essential genes and generates novel DNA sequences up to 96K base pairs.
Area of Science:
- Genomics
- Bioinformatics
- Computational Biology
Background:
- Large language models (LLMs) have shown success in various domains.
- Genomic data presents unique challenges for modeling due to its long and complex nature.
Purpose of the Study:
- To develop a novel, long-context generative model for genomic sequences.
- To leverage transformer architecture for understanding and generating DNA.
Main Methods:
- Developed a multiscale transformer model named megaDNA.
- Pre-trained the model on unannotated bacteriophage genomes using nucleotide-level tokenization.
Main Results:
- Demonstrated foundational capabilities including predicting essential genes, genetic variant effects, regulatory element activity, and taxonomy.
- Successfully generated de novo DNA sequences up to 96K base pairs.
- Generated sequences contained potential regulatory elements and phage-related proteins.
Conclusions:
- megaDNA represents a significant advancement in generative genomic modeling.
- The model shows potential for diverse applications in genomic analysis and synthetic biology.
Related Concept Videos
Lytic Cycle of Bacteriophages
Lysogenic Cycle of Bacteriophages
Viral Replication: Lysogenic Cycle
Viral Replication: Lytic Cycle
Viruses with RNA Genomes
Leaky Scanning

