Unsupervised AI reveals insect species-specific genome signatures

Area of Science:

• Comparative genomics
• Bioinformatics
• Evolutionary biology

Background:

• Insect genomes exhibit high diversity in traits like wings and metamorphosis, making them valuable for studying genome evolution.
• Comparative genomic studies have explored various insect phylogenetic ranges, but understanding species-specific genomic characteristics remains crucial.
• Oligonucleotide composition, or genome signatures, offers insights into species-specific genomic differences and their biological significance.

Purpose of the Study:

• To characterize species-specific oligonucleotide compositions (genome signatures) in a wide phylogenetic range of 22 insect species.
• To identify specific genomic regions with distinct oligonucleotide compositions within insect genomes.
• To compare insect genome characteristics with those of vertebrates, particularly humans, using machine learning approaches.

Main Methods:

• Analysis of genomic fragments (100-kb or 1-Mb sequences) from 22 insect species using a batch-learning self-organizing map (BLSOM).
• BLSOM, an unsupervised machine learning algorithm, was employed to extract species-specific oligonucleotide compositions (genome signatures).
• Clustering of sequences based solely on oligonucleotide composition, enabling both interspecies and intraspecies separation.

Main Results:

• Successfully characterized species-specific genome signatures across a wide phylogenetic range of insects.
• Identified distinct genomic regions with unique oligonucleotide compositions, such as Mb-level structures in the grasshopper *Schistocerca americana*.
• Observed similarities in distinct Mb-length genomic regions between insects and humans, suggesting conserved genome organization principles.

Conclusions:

• Genome signatures derived from oligonucleotide composition are effective for distinguishing insect species and identifying unique genomic regions.
• The findings highlight the potential of genome signatures as a tool to explore non-coding DNA functions and uncover genomic mysteries.
• Comparing insect and human genomes using this approach provides a framework for understanding genome evolution across distant taxa.