Identifying Black Corals and Octocorals From Deep-Sea Imagery for Ecological Assessments: Trade-Offs Between Morphology and Taxonomy

An increased reliance on imagery as the source of biodiversity data from the deep sea has stimulated many recent advances in image annotation and data management. The form of image-derived data is determined by the way faunal units are classified and should align with the needs of the ecological study to which it is applied. Some applications may require only low-resolution biodiversity data, which is easier and cheaper to generate, whereas others will require well-resolved biodiversity measures, which require a larger investment in annotation methods. We assessed these trade-offs using a dataset of 5 939 images and physical collections of black and octocorals taken during surveys from a seamount area in the southwest Pacific Ocean. Coral diversity was greatly underestimated in images: only 55 black and octocoral ‘phototaxa’ (best-possible identifications) were consistently distinguishable out of a known 210 species in the region (26%). Patterns of assemblage composition were compared between the phototaxa and a standardized Australian classification scheme (“CATAMI”) that uses morphotypes to classify taxa. Results were similar in many respects, but the identities of dominant, and detection of rare but locally abundant, coral entities were achieved only when annotation was at phototaxon resolution, and when faunal densities were recorded. A case study of data from 4 seamounts compared three additional classification schemes. Only the two with highest resolution – phototaxon and a combined phototaxon-morphological scheme – were able to distinguish black and octocoral communities on unimpacted vs. impacted seamounts. We conclude that image annotation schemes need to be fit-for-purpose. Morphological schemes such as CATAMI may perform well and are most easily standardized for cross-study data sharing, but high resolution (and more costly) annotation schemes are likely necessary for some ecological and management-based applications including biodiversity inventory, change detection (monitoring) – and to develop automated annotation using machine learning.

HORmon: automated annotation of human centromeres

Recent advances in long-read sequencing opened a possibility to address the long-standing questions about the architecture and evolution of human centromeres. They also emphasized the need for centromere annotation (partitioning human centromeres into monomers and higher-order repeats (HORs)). Even though there was a half-century-long series of semi-manual studies of centromere architecture, a rigorous centromere annotation algorithm is still lacking. Moreover, an automated centromere annotation is a prerequisite for studies of genetic diseases associated with centromeres, and evolutionary studies of centromeres across multiple species. Although the monomer decomposition (transforming a centromere into a monocentromere written in the monomer alphabet) and the HOR decomposition (representing a monocentromere in the alphabet of HORs) are currently viewed as two separate problems, we demonstrate that they should be integrated into a single framework in such a way that HOR (monomer) inference affects monomer (HOR) inference. We thus developed the HORmon algorithm that integrates the monomer/HOR inference and automatically generates the human monomers/HORs that are largely consistent with the previous semi-manual inference.

Somnotate: A robust automated sleep stage classifier that exceeds human performance and identifies ambiguous states in mice

Manual sleep stage annotation is a time-consuming but often essential step in the analysis of sleep data. To address this bottleneck several algorithms have been proposed that automate this process, reporting performance levels that are on par with manual annotation according to measures of inter-rater agreement. Here we first demonstrate that inter-rater agreement can provide a biased and imprecise measure of annotation quality. We therefore develop a principled framework for assessing performance against a consensus annotation derived from multiple experienced sleep researchers. We then construct a new sleep stage classifier that combines automated feature extraction using linear discriminant analysis with inference based on vigilance state-dependent contextual information using a hidden Markov model. This produces automated annotation accuracies that exceed expert performance on rodent electrophysiological data. Furthermore, our classifier is shown to be robust to errors in the training data, robust to experimental manipulations, and compatible with different recording configurations. Finally, we demonstrate that the classifier identifies both successful and failed attempts to transition between vigilance states, which may offer new insights into the occurrence of short awake periods between REM and NREM sleep. We call our classifier 'Somnotate' and make an implementation available to the neuroscience community.

Robots Versus Humans: Automated Annotation Accurately Quantifies Essential Ocean Variables of Rocky Intertidal Functional Groups and Habitat State

Standardized methods for effectively and rapidly monitoring changes in the biodiversity of marine ecosystems are critical to assess status and trends in ways that are comparable between locations and over time. In intertidal and subtidal habitats, estimates of fractional cover and abundance of organisms are typically obtained with traditional quadrat-based methods, and collection of photoquadrat imagery is a standard practice. However, visual analysis of quadrats, either in the field or from photographs, can be very time-consuming. Cutting-edge machine learning tools are now being used to annotate species records from photoquadrat imagery automatically, significantly reducing processing time of image collections. However, it is not always clear whether information is lost, and if so to what degree, using automated approaches. In this study, we compared results from visual quadrats versus automated photoquadrat assessments of macroalgae and sessile organisms on rocky shores across the American continent, from Patagonia (Argentina), Galapagos Islands (Ecuador), Gorgona Island (Colombian Pacific), and the northeast coast of the United States (Gulf of Maine) using the automated software CoralNet. Photoquadrat imagery was collected at the same time as visual surveys following a protocol implemented across the Americas by the Marine Biodiversity Observation Network (MBON) Pole to Pole of the Americas program. Our results show that photoquadrat machine learning annotations can estimate percent cover levels of intertidal benthic cover categories and functional groups (algae, bare substrate, and invertebrate cover) nearly identical to those from visual quadrat analysis. We found no statistical differences of cover estimations of dominant groups in photoquadrat images annotated by humans and those processed in CoralNet (binomial generalized linear mixed model or GLMM). Differences between these analyses were not significant, resulting in a Bray-Curtis average distance of 0.13 (sd 0.11) for the full label set, and 0.12 (sd 0.14) for functional groups. This is the first time that CoralNet automated annotation software has been used to monitor “Invertebrate Abundance and Distribution” and “Macroalgal Canopy Cover and Composition” Essential Ocean Variables (EOVs) in intertidal habitats. We recommend its use for rapid, continuous surveys over expanded geographical scales and monitoring of intertidal areas globally.

Bias-invariant RNA-sequencing metadata annotation

Background Recent technological advances have resulted in an unprecedented increase in publicly available biomedical data, yet the reuse of the data is often precluded by experimental bias and a lack of annotation depth and consistency. Missing annotations makes it impossible for researchers to find datasets specific to their needs. Findings Here, we investigate RNA-sequencing metadata prediction based on gene expression values. We present a deep-learning–based domain adaptation algorithm for the automatic annotation of RNA-sequencing metadata. We show, in multiple experiments, that our model is better at integrating heterogeneous training data compared with existing linear regression–based approaches, resulting in improved tissue type classification. By using a model architecture similar to Siamese networks, the algorithm can learn biases from datasets with few samples. Conclusion Using our novel domain adaptation approach, we achieved metadata annotation accuracies up to 15.7% better than a previously published method. Using the best model, we provide a list of >10,000 novel tissue and sex label annotations for 8,495 unique SRA samples. Our approach has the potential to revive idle datasets by automated annotation making them more searchable.

Fast Machine Learning Annotation in the Medical Domain: A Semi-Automated Video Annotation Tool for Gastroenterologists

Background: Machine learning, especially deep learning, is becoming more and more relevant in research and development in the medical domain. For all of the supervised deep learning applications, data is the most critical factor in securing successful implementation and sustaining the progress of the machine learning model. Especially gastroenterological data, which often involves endoscopic videos, are cumbersome to annotate. Domain experts are needed to interpret and annotate the videos. To support those domain experts, we generated a framework. With this framework, instead of annotating every frame in the video sequence, experts are just performing key annotations at the beginning and the end of sequences with pathologies, e.g. visible polyps. Subsequently, non-expert annotators supported by machine learning add the missing annotations for the frames in-between. Results: Using this framework we were able to reduce work load of domain experts on average by a factor of 20. This is primarily due to the structure of the framework, which is designed to minimize the workload of the domain expert. Pairing this framework with a state-of-the-art semi-automated pre-annotation model enhances the annotation speed further. Through a study with 10 participants we show that semi-automated annotation using our tool doubles the annotation speed of non-expert annotators compared to a well-known state-of-the-art annotation tool. Conclusion: In summary, we introduce a framework for fast expert annotation for gastroenterologists, which reduces the workload of the domain expert considerably while maintaining a very high annotation quality. The framework incorporates a semi-automated annotation system utilizing trained object detection models. The software and framework are open-source.