Discovering Unknown Insects With eDNA and AI

Biodiversity researchers are rapidly accelerating the discovery of unknown insect species by combining environmental DNA (eDNA) metabarcoding with artificial intelligence (AI). This article provides an overview of how scientists collect genetic fragments from the environment, sequence them in bulk, and use machine learning algorithms to identify and catalog previously unknown insects without ever needing to capture a physical specimen.

Insects represent the largest group of animals on Earth, yet millions of species remain unclassified. Traditional methods of cataloging insects require trapping specimens and having human experts manually examine their physical traits under a microscope. This process is slow, expensive, and struggles to keep pace with global habitat loss. To overcome this bottleneck, scientists now turn to eDNA and AI.

Every insect leaves behind microscopic traces of its DNA in its habitat through shed skin, waste, saliva, or decaying body parts. Researchers collect these genetic traces by gathering samples from soil, water, air filters, or even rainwater washed off tree leaves. Instead of isolating individual insects, scientists extract all the genetic material present in a single environmental sample at once.

Once the DNA is extracted, researchers use a technique called metabarcoding. They target a specific, short section of a gene—commonly the cytochrome c oxidase I (COI) gene in insects—which acts as a biological barcode. High-throughput sequencing machines then read millions of these DNA barcodes simultaneously from the mixed sample. This produces a massive dataset representing every insect species that recently interacted with that environment.

The biggest challenge with metabarcoding is the "dark taxa" problem. When researchers compare their newly sequenced DNA barcodes against global reference libraries, a vast majority of the sequences do not match any known species. Because the species have never been formally described by science, traditional database searches hit a dead end.

This is where artificial intelligence transforms the process. Researchers feed the millions of unidentified DNA sequences into advanced machine learning models. The AI begins by filtering out sequencing errors and genetic noise, ensuring that researchers do not mistake corrupted data for a new species.

Next, unsupervised machine learning algorithms group the clean, unknown DNA sequences based on subtle patterns and similarities in their genetic code. These groups are called Operational Taxonomic Units (OTUs), which serve as digital stand-ins for distinct species. Deep learning models, trained on the DNA of known insects, then analyze these unknown clusters to predict where they fit on the evolutionary tree. Even if an insect is completely new to science, the AI can accurately determine its family or genus by recognizing deep evolutionary patterns in the genetic code.

Furthermore, AI combines this genetic data with environmental metadata, such as the specific location, climate, and vegetation where the sample was taken. By cross-referencing genetic clusters with ecological data, the AI helps researchers understand the habitat preferences and geographical range of these newly discovered, unseen insects. This integration of eDNA metabarcoding and AI allows scientists to map and monitor Earth's hidden insect biodiversity in a fraction of the time required by traditional methods.