Reading the Water
An ocean-wide genetic archive — and the tools finally capable of opening it.
The census of deep-sea life has always been a study in epistemological humility. We know more about the surface of Mars than we do about the ocean floor. The hadal zone — that vast, lightless territory below 6,000 metres, comprising the world's submarine trenches — covers an area roughly equivalent to Australia, and until recently was assumed to be biologically sparse. That assumption is now collapsing, not because the deep has changed, but because the tools we bring to it finally have.
Environmental DNA — eDNA — is the molecular residue organisms leave in their surroundings: shed cells, mucus, faeces, metabolic by-products. In theory, a volume of seawater is a biological archive. In practice, extracting meaning from that archive has required the maturation of an entire analytical pipeline: sample collection at depth, preservation, filtration through membranes as fine as 0.22 micrometres, high-throughput sequencing, and — the critical step — comparison against reference libraries that themselves require years to build.
The convergence of eDNA metabarcoding and deep-learning taxonomy systems is producing what one researcher called "a second Challenger expedition." The first, launched in 1872, sent naturalists around the globe over four years, yielding 4,700 new species. The second is happening in laboratories, in supercomputing clusters, in water samples collected by autonomous underwater vehicles. It is faster, cheaper, and almost incomprehensibly more thorough.
A 2024 study published in Communications Biology used sponge tissue from four biogeographic regions of the North Atlantic deep sea — at depths between 40 and 2,750 metres — to recover eDNA from the surrounding benthic communities. The sponges, passive filter-feeders, had absorbed years of biological signal from their environment. Screening 97 sponge tissues using the universal COI barcode region, the researchers recovered at least 406 metazoan species, with particular diversity across Chordata, Cnidaria, Echinodermata, and Porifera. The phrase used was "unprecedented levels of taxonomic diversity per unit effort." A sponge had done what years of trawling could not.
Names We Do Not Yet Have
How a bottleneck of expertise is being broken open — workshop by workshop, species by species.
Deep-sea taxonomy has long suffered from what researchers politely call a "bottleneck." The species are collected — aboard research vessels, by submersibles, by baited lander traps — but the specialists trained to formally describe and name them are few. Describing a new species is slow, meticulous work: morphological examination, comparison against type specimens in museum collections, consultation of literature reaching back to the Victorian era, formal publication. The International Seabed Authority recognised the scale of the problem in 2022, convening a dedicated workshop on enhancing genetic approaches to advance deep-sea taxonomy. The result was the "One Thousand Reasons" project — an initiative to formally describe 1,000 new deep-sea species by the end of the decade.
In 2026, a week-long taxonomy workshop involving 16 specialists described 24 new amphipod species from the Clarion-Clipperton Zone — including a new family, Mirabestiidae, and superfamily Mirabestioidea: entirely new branches on the tree of life.
Jażdżewska & Horton et al., National Oceanography Centre, 2026North Atlantic sponge tissue used as a natural eDNA sampler uncovered unprecedented diversity at depths between 40 and 2,750 metres — spanning Chordata, Cnidaria, Echinodermata, and Porifera. A sponge did what years of trawling could not.
Wan et al., Communications Biology, 2024Mitochondrial DNA analysis collapsed the isolation model — species thought to be trench-specific proved to range across the Pacific and Indian Oceans. The genus Notoliparis turned out synonymous with Pseudoliparis.
Global Deep-Sea Trenches Exploration Program, 2025Interpretable prototype-based CNNs using ProtoPNet now outpace traditional bioinformatic methods, while visualising which base pairs drive each classification — opening the black box to expert scrutiny.
Scientific Reports, 2025Each new species description constitutes a data point in a map of planetary biodiversity that directly shapes conservation decisions, mining regulations, and our understanding of how evolution operates under extremity. Without formal description, a creature has no legal protection, no place in the databases that AI systems train against, no existence in the scientific record. Taxonomy is not administrative housekeeping — it is the infrastructure on which all downstream biology depends.
Trenches as Corridors
A century-old isolation model — overturned by a single gene mutation at 7,000 metres.
Perhaps the most philosophically disorienting finding of the new deep-sea genetics has been the collapse of isolation as a model. For decades, researchers assumed that hadal trenches — separated by thousands of kilometres of abyssal plain — functioned as evolutionary islands. Deep species were thought to be discrete, locally adapted, reproductively isolated by distance and depth.
A 2025 study in Cell, conducted under China's Global Deep-Sea Trenches Exploration Program, confirmed two distinct pathways by which vertebrates colonised the deep: ancient lineages already adapted to cold, high-pressure environments, and newer immigrants descending from shallower waters through rapid physiological adjustment. The rtf1 mutation identified in hadal snailfish, aiding protein stability under crushing pressure, exemplifies the second route — evolution working fast, at depth, through single genetic changes with large phenotypic consequences.
The first scientific deep-sea expedition circles the globe over four years, recovering 4,700 new species using weighted ropes and dredge nets. The gold standard for a century.
North Atlantic study demonstrates sponge tissue passively captures eDNA from surrounding benthic communities — recovering 406 metazoan species from 97 samples. A new monitoring paradigm emerges.
A Cell study on 11 deep-sea fish species from 1,218–7,730m identifies a single rtf1 gene mutation enabling pressure adaptation. PCBs detected in Mariana Trench snailfish confirm abyssal pollution penetration.
In one week, 16 specialists describe 24 new amphipod species — including an entirely new family and superfamily, Mirabestiidae and Mirabestioidea. The One Thousand Reasons project accelerates.
The same study detected elevated concentrations of polychlorinated biphenyls — PCBs, synthetic industrial pollutants banned in many countries since the 1970s — in the liver tissues of hadal snailfish from the Mariana Trench and the Philippine Sea Basin. The abyss is not sealed. What we have released into the upper ocean does not stay there. The deepest, most remote organisms on the planet carry the chemical signature of human industrial activity in their tissues.
Algorithms in the Dark
A tug of war between discovery and the limits of what machines can see — and what happens when they misidentify life.
The application of machine learning to deep-sea taxonomy sits at an uneasy frontier. Deep learning models trained on curated datasets — like MBARI's FathomNet database, drawing on nearly 26,000 hours of archival deep-sea video and 6.5 million annotated observations — can process footage orders of magnitude faster than human analysts. Convolutional neural networks have demonstrated speed advantages of up to 150 times over traditional bioinformatic methods. But the limits are significant.
A sequence belonging to an undescribed species — a creature for which there is no entry in any reference database — returns no useful classification. The model signals uncertainty, or worse, assigns the unknown to the nearest known neighbour, producing a confident misidentification. This is the deep irony of AI-assisted deep-sea taxonomy: the technology is most powerful precisely in the regions where the reference data is thinnest, and its errors are hardest to detect.
None of this occurs in a political vacuum. The Clarion-Clipperton Zone — the mid-Pacific abyssal plain where the most intensive deep-sea mining exploration is now underway — contains polymetallic nodule fields of enormous commercial value, and also, as recent surveys confirm, ecosystems of extraordinary biological richness. Europe's €9 million MiningImpact3 project is developing "traffic light" threshold systems — triggered by AI analysis of benthic imagery — designed to halt mining operations when ecological damage reaches critical levels.
The Second Expedition
What the new tools have produced, above all, is a precise measurement of how little we know.
In the spring of 2024, high school students in San Diego conducted the first student-to-student videoconference between a classroom and a research vessel operating at 4,500 metres depth. Through a satellite uplink, they watched ROV footage of the seafloor in real time and asked questions directly to the scientists piloting the vehicle. It was, by any measure, a minor technical achievement. As a symbol, it is something else: the deep ocean, for most of human history a place of inference and speculation, entering ordinary attention.
The scale of what remains unknown is genuinely staggering. Researchers now describe approximately 2,000 new ocean species per year — and the consensus among marine biologists is that this represents a small fraction of what exists. The eDNA surveys, the AI classifiers, the taxonomy workshops, the submersible expeditions: these are not converging on a complete inventory. They are establishing, for the first time, how large the inventory is. That knowledge — the knowledge of our ignorance, precisely measured — may be the most important thing the new tools have yet produced.
Speciest has always been, at one level, a meditation on what it means to share a planet with creatures we have not named, have not noticed, and in some cases will never encounter directly. The abyssal fauna — living at pressures that would collapse the human chest, in darkness that has persisted for geological eons — are the outer limit of that meditation. DNA and machine intelligence are not bringing them closer to us. They are revealing how far they always were.
Han Xu et al. "Evolution and genetic adaptation of fishes to the deep sea." Cell (2025) · Wan et al. "North Atlantic deep-sea benthic biodiversity unveiled through sponge natural sampler DNA." Communications Biology (2024) · Jażdżewska & Horton et al. "24 new species from the Clarion-Clipperton Zone." National Oceanography Centre (2026) · International Seabed Authority: "Environmental DNA studies have the potential to advance deep-sea biodiversity knowledge." Policy Brief (2024)
Buy me a coffee