BarcodeBERT: transformers for biodiversity analyses

Context Acarina are commonly collected in macroinvertebrate surveys used to monitor freshwater ecosystems. However, they can be difficult to identify morphologically requiring considerable taxonomic skill for identification to finer taxonomic levels. Therefore, in biomonitoring they are identified to subclass despite high species diversity and varied environmental responses. DNA barcoding individuals and DNA metabarcoding of bulk samples enables species to be accurately and routinely identified. However, poor DNA barcode coverage of Australian aquatic mites has hampered their use in DNA studies. Aims Here, we aim to generate DNA barcodes for mites from Greater Melbourne, Australia. Key results For many specimens, we link DNA barcodes to genus-level morphological identifications using genetic analysis of DNA barcodes to understand biodiversity. We then test if new DNA barcodes can improve identification of mites in samples processed with DNA metabarcoding. We found Australian aquatic mites showed high diversity with many DNA barcodes represented by single specimens. Conclusions Increased mite DNA barcode library coverage improved their detection using DNA metabarcoding. Implications Given high species diversity, much effort will be required to improve DNA barcode coverage for aquatic mites in Australia and integrate barcodes with species level taxonomy, allowing Acarina to be better incorporated into DNA-based biological monitoring.

Nuclear mitochondrial pseudogenes (numts) are nonfunctional copies of mtDNA in the nucleus that have been found in major clades of eukaryotic organisms. They can be easily coamplified with orthologous mtDNA by using conserved universal primers; however, this is especially problematic for DNA barcoding, which attempts to characterize all living organisms by using a short fragment of the mitochondrial cytochrome c oxidase I (COI) gene. Here, we study the effect of numts on DNA barcoding based on phylogenetic and barcoding analyses of numt and mtDNA sequences in two divergent lineages of arthropods: grasshoppers and crayfish. Single individuals from both organisms have numts of the COI gene, many of which are highly divergent from orthologous mtDNA sequences, and DNA barcoding analysis incorrectly overestimates the number of unique species based on the standard metric of 3% sequence divergence. Removal of numts based on a careful examination of sequence characteristics, including indels, in-frame stop codons, and nucleotide composition, drastically reduces the incorrect inferences of the number of unique species, but even such rigorous quality control measures fail to identify certain numts. We also show that the distribution of numts is lineage-specific and the presence of numts cannot be known a priori. Whereas DNA barcoding strives for rapid and inexpensive generation of molecular species tags, we demonstrate that the presence of COI numts makes this goal difficult to achieve when numts are prevalent and can introduce serious ambiguity into DNA barcoding.

BackgroundThe key idea of DNA barcode initiative is to identify, for each group of species belonging to different kingdoms of life, a short DNA sequence that can act as a true taxon barcode. DNA barcode represents a valuable type of information that can be integrated with ecological, genetic, and morphological data in order to obtain a more consistent taxonomy. Recent studies have shown that, for the animal kingdom, the mitochondrial gene cytochrome c oxidase I (COI), about 650 bp long, can be used as a barcode sequence for identification and taxonomic purposes of animals. In the present work we aims at introducing the use of an alignment-free approach in order to make taxonomic analysis of barcode sequences. Our approach is based on the use of two compression-based versions of non-computable Universal Similarity Metric (USM) class of distances. Our purpose is to justify the employ of USM also for the analysis of short DNA barcode sequences, showing how USM is able to correctly extract taxonomic information among those kind of sequences.ResultsWe downloaded from Barcode of Life Data System (BOLD) database 30 datasets of barcode sequences belonging to different animal species. We built phylogenetic trees of every dataset, according to compression-based and classic evolutionary methods, and compared them in terms of topology preservation. In the experimental tests, we obtained scores with a percentage of similarity between evolutionary and compression-based trees between 80% and 100% for the most of datasets (94%). Moreover we carried out experimental tests using simulated barcode datasets composed of 100, 150, 200 and 500 sequences, each simulation replicated 25-fold. In this case, mean similarity scores between evolutionary and compression-based trees span between 83% and 99% for all simulated datasets.ConclusionsIn the present work we aims at introducing the use of an alignment-free approach in order to make taxonomic analysis of barcode sequences. Our approach is based on the use of two compression-based versions of non-computable Universal Similarity Metric (USM) class of distances. This way we demonstrate the reliability of compression-based methods even for the analysis of short barcode sequences. Compression-based methods, with their strong theoretical assumptions, may then represent a valid alignment-free and parameter-free approach for barcode studies.

Compared to traditional morphological identification, DNA barcoding-molecular identification based on sequencing of a segment of mitochondrial cytochrome c oxidase subunit I (COI)-provides a shortcut to authenticating chelonian identifications. Here, we selected 63 government-seized chelonian specimens deposited at Taipei Zoo for DNA barcoding analysis. DNA barcoding and subsequent phylogenetic analysis successfully authenticated 36 chelonian species, including five that are listed in CITES Appendix I. Approximately 90% (57/63) of the specimens were successfully authenticated by our molecular approach, but lack or error of BOLD reference sequences, biological processes such as hybridization, and uncertain species delimitation all reduced the accuracy of DNA barcoding. To increase the accuracy of DNA barcoding, Taipei Zoo will continue to enrich the BOLD database and also establish a genetic database, to include additional genetic markers, by using government-seized chelonian specimens. A fast and accurate method to authenticate seized samples could assist law enforcement agencies to prosecute criminals and restrict illegal exploitation of wild chelonian resources.

DNA technologies have many advantages for biomonitoring and biodiversity analyses, but these depend on the availability of relevant reference DNA barcodes. To be most useful, a DNA barcode should be linked to a taxonomic name, which can in turn be connected to ecological information. This linking can be achieved by DNA barcoding of taxonomically identified specimens. Museums are a promising source of such specimens, but the DNA in museum specimens is often degraded, necessitating carefully optimised DNA extraction methods. In this issue of Molecular Ecology Resources, Holmquist etal. (2025) present a DNA extraction protocol for museum insect specimens, using in-house formulated Solid Phase Reversible Immobilisation (SPRI) beads. The authors carried out several experiments with statistical evaluation to determine optimal DNA extraction parameters, before testing the protocol on a large and diverse pool of museum-held insect specimens. The result is a low-cost and effective DNA extraction protocol for diverse museum insect specimens.

Federated learning (FL) has arisen as an alternative to collecting large amounts of data in a central place to train a machine learning (ML) model. FL is privacy-friendly, allowing multiple parties to collaboratively train an ML model without exchanging or transmitting their training data. For this purpose, an aggregator iteratively coordinates the training process among parties, and parties simply share with the aggregator model updates, which contain information pertinent to the model such as neural network weights. Besides privacy, generalization has been another key driver for FL: parties who do not have enough data to train a good performing model by themselves can now engage in FL to obtain an ML model suitable for their tasks. Products and real applications in the industry and consumer space have demonstrated the power of this learning paradigm. Recently, foundation models have taken the AI community by storm, promising to solve the shortage of labeled data. A foundation model is a powerful model that can be recycled for a variety of use cases by applying techniques such as zero-shot learning and full or parameter-efficient fine tuning. The premise is that the amount of data required to fine tune a foundation model for a new task is much smaller than fully training a traditional model from scratch. The reason why this is the case is that a good foundation model has already learned relevant general representations, and thus, adapting it to a new task only requires a minimal number of additional samples. This raises the question: Is FL still alive in the era of foundation models? In this talk, I will address this question. I will present some use cases where FL is very much alive. In these use cases, finding a foundation model with a desired representation is difficult if not impossible. With this pragmatic point of view, I hope to shed some light into a real use case where disparate private data is available in isolation at different parties and where labels may be located at a single party that doesn’t have any other information, making it impossible for a single party to train a model on its own. Furthermore, in some vertically-partitioned scenarios, cleaning data is not an option due to privacy-related reasons and it is not clear how to apply foundation models. Finally, I will also go over a few other requirements that are often overlooked, such as unlearning of data and its implications for the lifecycle management of FL and systems based on foundation models.

Acanthocephalans are multi-host endoparasites, many of which use freshwater amphipods as intermediate hosts for their larval stages (e.g., cystacanths) while adults live in the intestines of vertebrates, including waterfowl. In central Alberta, Canada, several co-occurring species of the acanthocephalan genus Polymorphus use the amphipod Gammarus lacustris Sars, 1863 as an intermediate host. We applied DNA barcoding and morphometric analysis to differentiate cystacanth larvae from G. lacustris sampled from 17 Albertan water bodies. We slide-mounted specimens and measured morphological traits relating to proboscis hooks. We sequenced the standard DNA barcoding region of the mitochondrial cytochrome c oxidase subunit I gene (COI). Morphometric analysis suggested that the acanthocephalans we collected belonged to four morphologically different groups that keyed to Polymorphus contortus (Bremser, 1821) Travassos, 1926; P. marilis Van Cleave, 1939; P. paradoxus Connel et Corner, 1957; and P. strumosoides (Lundström, 1942) Amin, 2013. Our Bayesian tree based on COI sequences generally corroborated the morphological results and supported that the specimens assigned to P. cf. contortus and P. cf. strumosoides belong to two distinct species. In contrast, the Bayesian tree showed that specimens of P. cf. marilis were nested as a cluster within the P. cf. paradoxus clade. Similarly, small pairwise genetic distance (< 2%) between specimens identified as P. cf. contortus and P. cf. strumosoides suggests that they are conspecific. Future studies should use morphology and sequence data from adult acanthocephalans to assess the taxonomic identity of the cystacanth-based Polymorphus taxa. Our study is the first to provide genetic information for the four Polymorphus taxa and emphasizes the importance of applying multiple approaches to differentiate parasite species.

Indoor CO2 forecasting supports proactive ventilation control that balances air quality with energy efficiency. While Machine Learning (ML) models have shown strong performance in controlled settings such as schools, their generalization across indoor spaces with diverse occupancy dynamics remains poorly characterized. We present a systematic benchmark of 11 forecasting models spanning simple baselines, statistical methods, classical ML, deep learning, ensembles, and foundation models using 18 weeks of IoT sensor data spanning six real-world use cases: conference rooms, dining halls, hospitals, food markets, offices and student residences. Performance depends strongly on the prediction horizon and on the regularity of occupancy-driven CO2 patterns. Simple baselines tend to perform best at short horizons (10 min ahead), while ensembles and fine-tuned foundation models provide more robust accuracy at longer horizons (4 h ahead). Remarkably, zero-shot foundation models demonstrate the ability to outperform trained classical models in data-scarce scenarios, challenging the traditional paradigm of localized training. These findings indicate that optimal forecasting strategies are context-dependent and challenge the assumption of universal model superiority.

In Washington D.C., recent legislation authorizes citizens to test if products are properly represented and, if they are not, to bring a lawsuit for the benefit of the general public. Recent studies revealing the widespread phenomenon of seafood substitution across the United States make it a fertile area for consumer protection testing. DNA barcoding provides an accurate and cost-effective way to perform these tests, especially when tissue alone is available making species identification based on morphology impossible. In this study, we sequenced the 5′ barcoding region of the Cytochrome Oxidase I gene for 12 samples of vertebrate and invertebrate food items across six restaurants in Washington, D.C. and used multiple analytical methods to make identifications. These samples included several ambiguous menu listings, sequences with little genetic variation among closely related species and one sequence with no available reference sequence. Despite these challenges, we were able to make identifications for all samples and found that 33% were potentially mislabeled. While we found a high degree of mislabeling, the errors involved closely related species and we did not identify egregious substitutions as have been found in other cities. This study highlights the efficacy of DNA barcoding and robust analyses in identifying seafood items for consumer protection.

DNA-based identification is vital for classifying biological specimens, yet methods to quantify the uncertainty of sequence-based taxonomic assignments are scarce. Challenges arise from noisy reference databases, including mislabelled entries and missing taxa. PROTAX addresses these issues with a probabilistic approach to taxonomic classification, advancing on methods that rely solely on sequence similarity. It provides calibrated probabilistic assignments to a partially populated taxonomic hierarchy, accounting for taxa that lack references and incorrect taxonomic annotation. While effective on smaller scales, global application of PROTAX necessitates substantially larger reference libraries, a goal previously hindered by computational barriers. We introduce PROTAX-GPU, a scalable algorithm capable of leveraging the global Barcode of Life Data System (>14 million specimens) as a reference database. Using graphics processing units (GPU) to accelerate similarity and nearest-neighbour operations and the JAX library for Python integration, we achieve over a 1000 × speedup compared with the central processing unit (CPU)-based implementation without compromising PROTAX's key benefits. PROTAX-GPU marks a significant stride towards real-time DNA barcoding, enabling quicker and more efficient species identification in environmental assessments. This capability opens up new avenues for real-time monitoring and analysis of biodiversity, advancing our ability to understand and respond to ecological dynamics. This article is part of the theme issue 'Towards a toolkit for global insect biodiversity monitoring'.

Chloroplast structure, codon usage bias, and machine learning-based molecular identification using DNA barcoding of Sophorae Tonkinensis Radix et Rhizoma (Shan Dou Gen) and its analogues.

The order Tetraodontiformes are one of the most unique groups of teleostean fish, exhibiting highly derived and greatly diversified phenotypes. It is a difficult task for both professionals and nonprofessionals to accurately identify these species only according to morphological characteristics. DNA barcoding can identify species at the molecular level to overcome the limitations of morphological classification. In this study, we collected 616 specimens of pufferfish from the coastal waters of China. According to the morphological characteristics, they were preliminarily identified as 50 species. Further analysis using DNA barcodes identified these specimens as 46 species, belonging to 23 genera, 6 families. According to the species classification results of DNA barcoding, the three species identified by morphology as Takifugu pseudommus, Takifugu chinensis, and Takifugu rubripes should be the same species. Similarly, Lagocephalus wheeleri is the synonym of Lagocephalus spadiceus. Another important discovery of DNA barcoding analysis is that there are closer interspecific genetic distances within the genus Takifugu. If T. rubripes, T. pseudommus, and T. chinensis are taken as one species, the average interspecific to intraspecific genetic distance ratio of Takifugu is only 6.21 times, which does not reach the DNA barcoding threshold of more than 10 times proposed previously. Although the interspecific genetic distance in the genus Takifugu is relatively small, each species can be clustered into independent clades in the NJ tree. In conclusion, this study not only found that there are synonymous phenomena in the order Tetraodontiformes but also provided molecular evidence for the valid species names of Takifugu rubripes and Lagocephalus Spadiceus. The results can provide reliable DNA barcoding information for the identification of pufferfish species, help solve the problem of classification confusion in this order, and provide technical support for the identification of the original components of related commodities on the aquatic product market.

Introduction: Traditional risk models predicting mortality in cardiogenic shock (CS) rely on static data snapshots, failing to capture the temporal dynamics represented in electronic health records. Conventional machine learning (ML) models can incorporate some temporal information but often struggle to model the complex, non-linear, and long-range relationships present in clinical time series data. Foundation models, pretrained on large, diverse time series datasets, offer a powerful alternative by capturing richer temporal patterns and making predictions in dynamic clinical settings. Hypothesis: We hypothesized that a ML approach using a pretrained multi-task time series model (UniTS) could be fine-tuned to predict mortality in patients with CS admitted to the cardiac intensive care unit (CICU) using high-resolution, multivariable clinical time series data. Methods: We performed a retrospective analysis of patients with CS admitted to the Brigham and Women’s Hospital CICU (2015-2024). Patients were split into training (80%) and validation (20%) cohorts. For each observation, 24h of clinical data were used to fine-tune a pretrained UniTS model to develop: (1) a dynamic model generating rolling 24h mortality predictions every 6h, and (2) a static model predicting overall in-hospital mortality after the first 24h of CICU admission. Validation performance was evaluated using area under the curve (AUC). The static model was compared to the IABP-Shock II risk score and SCAI shock stages for the same task using the DeLong test. Results: Among 2,109 admissions with CS (median age 68 years, 38% women), 25% were AMI-related, and 39% received temporary mechanical circulatory support. In-hospital mortality was 37%, with a median time to death of 3.0 days following CICU admission. The final model included 31 clinical variables ( Fig A ). The dynamic model achieved a per-prediction AUC of 0.88 for 24-hour mortality. The static model achieved an AUC of 0.83 for in-hospital mortality, significantly outperforming the IABP-Shock II risk score (AUC 0.74; p = 0.03) and SCAI Shock stage (AUC 0.61; p &lt; 0.001) ( Fig B ). Conclusion: Leveraging the full temporal and multivariable complexity of CICU time series clinical data, fine-tuned ML foundation models accurately predict both very early and in-hospital mortality. By substantially outperforming traditional risk stratification methods, this time series modeling approach offers a promising tool for dynamic risk assessment in CS.

Invertebrate biodiversity remains poorly understood although it comprises much of the terrestrial animal biomass, most species and supplies many ecosystem services. The main obstacle is specimen-rich samples obtained with quantitative sampling techniques (e.g., Malaise trapping). Traditional sorting requires manual handling, while molecular techniques based on metabarcoding lose the association between individual specimens and sequences and thus struggle with obtaining precise abundance information. Here we present a sorting robot that prepares specimens from bulk samples for barcoding. It detects, images and measures individual specimens from a sample and then moves them into the wells of a 96-well microplate. We show that the images can be used to train convolutional neural networks (CNNs) that are capable of assigning the specimens to 14 insect taxa (usually families) that are particularly common in Malaise trap samples. The average assignment precision for all taxa is 91.4% (75%-100%). This ability of the robot to identify common taxa then allows for taxon-specific subsampling, because the robot can be instructed to only pick a prespecified number of specimens for abundant taxa. To obtain biomass information, the images are also used to measure specimen length and estimate body volume. We outline how the DiversityScanner can be a key component for tackling and monitoring invertebrate diversity by combining molecular and morphological tools: the images generated by the robot become training images for machine learning once they are labelled with taxonomic information from DNA barcodes. We suggest that a combination of automation, machine learning and DNA barcoding has the potential to tackle invertebrate diversity at an unprecedented scale.

Machine learning has significantly advanced healthcare by aiding in disease prevention and treatment identification. However, accessing patient data can be challenging due to privacy concerns and strict regulations. Generating synthetic, realistic data offers a potential solution for overcoming these limitations, and recent studies suggest that fine-tuning foundation models can produce such data effectively. In this study, we explore the potential of foundation models for generating realistic medical images, particularly chest x-rays, and assess how their performance improves with fine-tuning. We propose using a Latent Diffusion Model, starting with a pre-trained foundation model and refining it through various configurations. Additionally, we performed experiments with input from a medical professional to assess the realism of the images produced by each trained model.