Subsampling Strategies Research Articles

Do Subsampling Strategies Reduce the Confounding Effect of Errors in Bispectral Retrievals on Estimates of Aerosol Cloud Interactions?

AbstractBi‐spectral retrievals of droplet effective radius and cloud optical depth are widely utilized to estimate aerosol cloud interactions (ACI) in warm clouds in the marine boundary layer. Here, we assess the effect of retrieval errors due to the neglect of 3D radiative transfer during the retrieval process on this analysis of ACI. We use an ensemble of stochastically‐modeled cloud fields and 3D radiative transfer simulations to study the retrieval errors at a solar zenith angle of 30°. Simulated retrieval biases in droplet number concentration (Nd) for all three MODIS channels vary systematically from +35% to −80% as cloud heterogeneity increases. Pixel‐level errors can be much larger. Commonly utilized subsampling strategies do not reduce the systematic variation in retrieval error. Negative error correlations between optical depth and droplet effective radius produce spuriously negative slopes between the logarithm of liquid water path and Nd (−1.0 to −0.3). Pixels at the center of (8 km)2 patches that are not overcast have a relative bias in Nd of −50%. The relative frequency of these biased pixels varies linearly with the clear fraction in MODIS data and form the basis for a simple parameterization of Nd bias that varies with cloud fraction (CF). Using this parameterization, synthetic data experiments indicate that estimates of the first indirect effect in the tropical ocean can be overestimated by up to 30%, and the effect of aerosol on CF is overestimated by 50%, due to neglecting the correlation between retrieval errors and cloud microphysics.

A non-monotone trust-region method with noisy oracles and additional sampling

In this work, we introduce a novel stochastic second-order method, within the framework of a non-monotone trust-region approach, for solving the unconstrained, nonlinear, and non-convex optimization problems arising in the training of deep neural networks. The proposed algorithm makes use of subsampling strategies that yield noisy approximations of the finite sum objective function and its gradient. We introduce an adaptive sample size strategy based on inexpensive additional sampling to control the resulting approximation error. Depending on the estimated progress of the algorithm, this can yield sample size scenarios ranging from mini-batch to full sample functions. We provide convergence analysis for all possible scenarios and show that the proposed method achieves almost sure convergence under standard assumptions for the trust-region framework. We report numerical experiments showing that the proposed algorithm outperforms its state-of-the-art counterpart in deep neural network training for image classification and regression tasks while requiring a significantly smaller number of gradient evaluations.

Advancing responsible genomic analyses of ancient mollusc shells.

The analysis of the DNA entrapped in ancient shells of molluscs has the potential to shed light on the evolution and ecology of this very diverse phylum. Ancient genomics could help reconstruct the responses of molluscs to past climate change, pollution, and human subsistence practices at unprecedented temporal resolutions. Applications are however still in their infancy, partly due to our limited knowledge of DNA preservation in calcium carbonate shells and the need for optimized methods for responsible genomic data generation. To improve ancient shell genomic analyses, we applied high-throughput DNA sequencing to 27 Mytilus mussel shells dated to ~111-6500 years Before Present, and investigated the impact, on DNA recovery, of shell imaging, DNA extraction protocols and shell sub-sampling strategies. First, we detected no quantitative or qualitative deleterious effect of micro-computed tomography for recording shell 3D morphological information prior to sub-sampling. Then, we showed that double-digestion and bleach treatment of shell powder prior to silica-based DNA extraction improves shell DNA recovery, also suggesting that DNA is protected in preservation niches within ancient shells. Finally, all layers that compose Mytilus shells, i.e., the nacreous (aragonite) and prismatic (calcite) carbonate layers, with or without the outer organic layer (periostracum) proved to be valuable DNA reservoirs, with aragonite appearing as the best substrate for genomic analyses. Our work contributes to the understanding of long-term molecular preservation in biominerals and we anticipate that resulting recommendations will be helpful for future efficient and responsible genomic analyses of ancient mollusc shells.

Unravelling bird nest arthropod community structure using metabarcoding

Bird nests are fascinating microcosms harboring a wide range of arthropods parasitizing the nesting birds or feeding on prey remains, feces, and the nest material. Studies of these communities have been entirely based on emergence traps which collect live organisms out of the nests. The analysis of nest contents and environmental DNA (eDNA) via metabarcoding could expand our knowledge and identify prey, exuviae, and other animal remains in bird nests. Here, we investigated the potential of arthropod remains, nest dust, and feathers to better describe arthropod diversity accumulated in 20 bird nests collected in Guelph (Canada). We used subsampling strategies and tested two extraction approaches to investigate the distribution of DNA in nests, account for low-quality DNA, and the presence of inhibitory substances. In total, 103 taxa were detected via metabarcoding. Arthropod remains delivered the highest number of taxa (n = 67), followed by nest dust (n = 29). Extractions with the PowerSoil kit outperformed DNeasy extractions coupled with PowerClean Pro inhibitor removal. Per nest, on average 5.5% taxonomic overlap between arthropod remains of different size classes was detected and subsamples of nest dust extracted with the PowerSoil kit showed 47.3% taxonomic overlap indicating a heterogeneous eDNA distribution in nests. Most detected species were either feeding in the nest, i.e., herbivorous / predatory, or bird food. We also detected molecular traces of 25 bird species, whose feathers were likely used as nest material. Consequently, the metabarcoding of bird nest materials provides a more complete picture of nest communities, which can enable future studies on functional diversity and better comparisons between nesting species.

Towards real-time STEM simulations through targeted subsampling strategies.

Scanning transmission electron microscopy images can be complex to interpret on the atomic scale as the contrast is sensitive to multiple factors such as sample thickness, composition, defects and aberrations. Simulations are commonly used to validate or interpret real experimental images, but they come at a cost of either long computation times or specialist hardware such as graphics processing units. Recent works in compressive sensing for experimental STEM images have shown that it is possible to significantly reduce the amount of acquired signal and still recover the full image without significant loss of image quality, and therefore it is proposed here that similar methods can be applied to STEM simulations. In this paper, we demonstrate a method that can significantly increase the efficiency of STEM simulations through a targeted sampling strategy, along with a new approach to independently subsample each frozen phonon layer. We show the effectiveness of this method by simulating a SrTiO3 grain boundary and monolayer 2H-MoS2 containing a sulphur vacancy using the abTEM software. We also show how this method is not limited to only traditional multislice methods, but also increases the speed of the PRISM simulation method. Furthermore, we discuss the possibility for STEM simulations to seed the acquisition of real data, to potentially lead the way to self-driving (correcting) STEM.

Investigation of radiomics based intra-patient inter-tumor heterogeneity and the impact of tumor subsampling strategies

While radiomics analysis has been applied for localized cancer disease, its application to the metastatic setting involves a non-exhaustive lesion subsampling strategy which may sidestep the intrapatient tumoral heterogeneity, hindering the reproducibility and the therapeutic response performance. Our aim was to evaluate if radiomics features can capture intertumoral intrapatient heterogeneity, and the impact of tumor subsampling on the computed heterogeneity. To this end, We delineated and extracted radiomics features of all visible tumors from single acquisition pre-treatment computed tomography of patients with metastatic lung cancer (cohort L) and confirmed our results on a larger cohort of patients with metastatic melanoma (cohort M). To quantify the captured heterogeneity, the absolute coefficient of variation (CV) of each radiomics index was calculated at the patient-level and a sensitivity analysis was performed using only a subset of all extracted features robust to the segmentation step. The extent of information loss by six commonly used tumor sampling strategies was then assessed. A total of 602 lesions were segmented from 43 patients (median age 57, 4.9% female). All robust radiomics indexes exhibited at least 20% of variation with significant heterogeneity both in heavily and oligo metastasized patients, and also at the organ level. None of the segmentation subsampling strategies were able to recover the true tumoral heterogeneity obtained by exhaustive tumor sampling. Image-based inter-tumor intra-patient heterogeneity can be successfully grasped by radiomics analyses. Failing to take into account this kind of heterogeneity will lead to inconsistent predictive algorithms. Guidelines to standardize the tumor sampling step and/or AI-driven tools to alleviate the segmentation effort are required.

CovSampler: A subsampling method with balanced genetic diversity for large-scale SARS-CoV-2 genome data sets.

Phylogenetic analysis has been widely used to describe, display and infer the evolutionary patterns of viruses. The unprecedented accumulation of SARS-CoV-2 genomes has provided valuable materials for the real-time study of SARS-CoV-2 evolution. However, the large number of SARS-CoV-2 genome sequences also poses great challenges for data analysis. Several methods for subsampling these large data sets have been introduced. However, current methods mainly focus on the spatiotemporal distribution of genomes without considering their genetic diversity, which might lead to postsubsampling bias. In this study, a subsampling method named covSampler was developed for the subsampling of SARS-CoV-2 genomes with consideration of both their spatiotemporal distribution and their genetic diversity. First, covSampler clusters all genomes according to their spatiotemporal distribution and genetic variation into groups that we call divergent pathways. Then, based on these divergent pathways, two kinds of subsampling strategies, representative subsampling and comprehensive subsampling, were provided with adjustable parameters to meet different users’ requirements. Our performance and validation tests indicate that covSampler is efficient and stable, with an abundance of options for user customization. Overall, our work has developed an easy-to-use tool and a webserver (https://www.covsampler.net) for the subsampling of SARS-CoV-2 genome sequences.

Spatial distribution of microplastics in the tropical Indian Ocean based on laser direct infrared imaging and microwave-assisted matrix digestion

Suspended particulate matter was collected from subsurface (6 m) water along an E-W transect through the tropical Indian Ocean using a specialized inert (plastic free) fractionated filtration system. The samples were subjected to a new microwave-assisted “one-pot” matrix removal (efficiency: 94.3% ± 0.3% (1 SD, n = 3)) and microplastic extraction protocol (recovery: 95% ± 4%). The protocol enables a contamination-minimized digestion and requires only four filtration steps. In comparison, classical sample processing approaches involve up to eight filtration steps until the final analysis. Microplastics were identified and physically characterized by means of a novel quantum cascade laser-based imaging routine.LDIR imaging facilitates the analysis of up to 1000 particles/fibers (<300 μm) within approximately 1–2 h. In comparison to FTIR and Raman imaging, it can help to circumvent uncertainties, e. g. from subsampling strategies due to long analysis and post-processing times of large datasets. Over 97% of all particles were correctly identified by the automated routine - without spectral reassignments. Moreover, 100% agreement was obtained between ATR-FTIR and LDIR-based analysis regarding particles and fibers >300 μm.The mean microplastic concentration of the analyzed samples was 50 ± 30 particles/fibers m−3 (1 SD, n = 21). Number concentrations ranged from 8 to 132 particles/fibers m−3 (20–300 μm). The most abundant polymer clusters were acrylates/polyurethane/varnish (49%), polyethylene terephthalate (26%), polypropylene (8%), polyethylene (4%) and ethylene-vinyl acetate (4%). 96% of the microplastic particles had a diameter <100 μm. Though inter-study comparison is difficult, the investigated area exhibits a high contamination with particulate plastics compared to other open ocean regions. A distinct spatial trend was observed with an increasing share of the size class 20–50 μm from east to west.

Combining farm and household surveys with modelling approaches to improve post-harvest loss estimates and reduce data collection costs

While there is growing awareness of the issue of food losses at the political level, official post-harvest loss data for informing policymaking and reporting on SDG Indicator 12.3.1. (a) Food Loss Index is scarce. Representative sample-based surveys are necessary to obtain information on on-farm losses at the country level, but due to the issue’s complexity, a loss module covering several key questions is needed. One main strategy proposed by the 50x2030 Initiative for optimizing data collection is sub-sampling for some of the survey modules. This paper examines whether modelling approaches can be combined with sub-sampling to improve harvest and post-harvest loss estimates and allow for further sample and cost reduction. The paper first presents the loss models generated on four selected surveys conducted in Malawi, Zimbabwe, and Nigeria, which were built using the Classification and Regression Tree (CART) method. The performance of each model is assessed for different sizes of sub-samples to improve the sample-based estimates, either by model-based estimates or by model-based imputation. The research concludes that the model-based estimates improve the loss estimates of the sub-samples due to post-stratification implied in the CART method, whereby they can constitute a cost-effective complement to sub-sampling strategies, while model-based imputations should only be used on a reduced number of missing observations. The models perform best when the survey invests in obtaining more detailed on-farm loss data and considers some key variables identified as relevant for on-farm loss models. Sub-sampling allows for investment in more detailed questionnaires and some considerations are derived for its design.

Modeling mean estimation tasks in within-trial and across-trial contexts.

The mean estimation task, which explicitly asks observers to estimate the mean feature value of multiple stimuli, is a fundamental paradigm in research areas such as ensemble coding and cue integration. The current study uses computational models to formalize how observers summarize information in mean estimation tasks. We compare model predictions from our Fidelity-based Integration Model (FIM) and other models on their ability to simulate observed patterns in within-trial weight distribution, across-trial information integration, and set-size effects on mean estimation accuracy. Experiments show non-equal weighting within trials in both sequential and simultaneous mean estimation tasks. Observers implicitly overestimated trial means below the global mean and underestimated trial means above the global mean. Mean estimation performance declined and stabilized with increasing set sizes. FIM successfully simulated all observed patterns, while other models failed. FIM's information sampling structure provides a new way to interpret the capacity limit in visual working memory and sub-sampling strategies. As a model framework, FIM offers task-dependent modeling for various ensemble coding paradigms, facilitating research synthesis across different studies in the literature.

Prioritization‐based subsampling quality assessment methodology for mobility‐related information systems

Mobility-related information systems, such as on-street parking information (OSPI) systems have become more popular in the original equipment manufacturer (OEM) industry over the last decade. However, there is a lack of methods to assess their quality at a large scale. This paper introduces a data-driven methodology to measure the true quality by fleet data prioritization-based subsampling strategies (PSSs). It is applied to the use case of OSPI using parking events (PE), but is applicable to other mobility-related information systems utilizing their respective fleet data. PSSs are defined based on neighbourhoods and time periods. Each PSS generates a unique set of spatio-temporally important areas at different quadkey zoom levels over 168 week-hours, called slices. The importance weight in each slice depends on the volume of PE within them. The algorithm for each PSS automatically selects important areas and time frames that are vital to be observed. Sample prediction models are used for the benefits assessment of the methodology by comparing it against non-prioritized randomized selection of ground truth. It is proven that the methodology can lessen the effort of ground truth collection, while maintaining the amount of information necessary to assess the true quality of a prediction model.

Sparsity-driven deterministic sampling strategy for coded aperture x-ray computed tomography

The sub-sampling strategies in x-ray computed tomography (CT) have gained importance due to their practical relevance. In this direction of research, also known as coded aperture x-ray computed tomography (CAXCT), both random and deterministic strategies have been proposed in the literature. Of the techniques available, those based on compressive sensing (CS) have recently gained more traction because CS-based ideas efficiently exploit inherent duplication present in the system. The quality of the reconstructed CT images nevertheless depends on the sparse signal recovery properties of the sub-sampled Radon matrices. In the present work, we optimize CAXCT deterministically in such a way that the corresponding sub-sampled Radon matrices remain close to the equiangular tight frames for better numerical behaviour. We show that this optimization, involving the Khatri–Rao product, leads to a non-negative sparse approximation. While comparing and contrasting our method with its existing counterparts, we show that the proposed algorithm is computationally less involved. Finally, we demonstrate efficacy of the proposed deterministic sub-sampling strategy in recovering CT images in both noiseless and noisy cases.

A Training Set Subsampling Strategy for the Reduced Basis Method

We present a subsampling strategy for the offline stage of the Reduced Basis Method. The approach is aimed at bringing down the considerable offline costs associated with using a finely-sampled training set. The proposed algorithm exploits the potential of the pivoted QR decomposition and the discrete empirical interpolation method to identify important parameter samples. It consists of two stages. In the first stage, we construct a low-fidelity approximation to the solution manifold over a fine training set. Then, for the available low-fidelity snapshots of the output variable, we apply the pivoted QR decomposition or the discrete empirical interpolation method to identify a set of sparse sampling locations in the parameter domain. These points reveal the structure of the parametric dependence of the output variable. The second stage proceeds with a subsampled training set containing a by far smaller number of parameters than the initial training set. Different subsampling strategies inspired from recent variants of the empirical interpolation method are also considered. Tests on benchmark examples justify the new approach and show its potential to substantially speed up the offline stage of the Reduced Basis Method, while generating reliable reduced-order models.

Frequency Subsampling of Ultrasound Nondestructive Measurements: Acquisition, Reconstruction, and Performance.

In ultrasound nondestructive testing (NDT), a widespread approach is to take synthetic aperture measurements from the surface of a specimen to detect and locate defects within it. Based on these measurements, imaging is usually performed using the synthetic aperture focusing technique (SAFT). However, SAFT is suboptimal in terms of resolution and requires oversampling in the time domain to obtain a fine grid for the delay-and-sum (DAS). On the other hand, parametric reconstruction algorithms give better resolution, but their usage for imaging becomes computationally expensive due to the size of the parameter space and a large amount of measurement data in realistic 3-D scenarios when using oversampling. In the literature, the remedies to this are twofold. First, the amount of measurement data can be reduced using state-of-the-art sub-Nyquist sampling approaches to measure Fourier coefficients instead of time-domain samples. Second, parametric reconstruction algorithms mostly rely on matrix-vector operations that can be implemented efficiently by exploiting the underlying structure of the model. In this article, we propose and compare different strategies to choose the Fourier coefficients to be measured. Their asymptotic performance is compared by numerically evaluating the Cramér-Rao bound (CRB) for the localizability of the defect coordinates. These subsampling strategies are then combined with an l1 -minimization scheme to compute 3-D reconstructions from the low-rate measurements. Compared to conventional DAS, this allows us to formulate a fully physically motivated forward model matrix. To enable this, the projection operations of the forward model matrix are implemented matrix-free by exploiting the underlying two-level Toeplitz structure. Finally, we show that high-resolution reconstructions from as low as a single Fourier coefficient per A-scan are possible based on simulated data and measurements from a steel specimen.

Phylodynamics reveals the role of human travel and contact tracing in controlling the first wave of COVID-19 in four island nations.

New Zealand, Australia, Iceland, and Taiwan all saw success in controlling their first waves of Coronavirus Disease 2019 (COVID-19). As islands, they make excellent case studies for exploring the effects of international travel and human movement on the spread of COVID-19. We employed a range of robust phylodynamic methods and genome subsampling strategies to infer the epidemiological history of Severe acute respiratory syndrome coronavirus 2 in these four countries. We compared these results to transmission clusters identified by the New Zealand Ministry of Health by contact tracing strategies. We estimated the effective reproduction number of COVID-19 as 1–1.4 during early stages of the pandemic and show that it declined below 1 as human movement was restricted. We also showed that this disease was introduced many times into each country and that introductions slowed down markedly following the reduction of international travel in mid-March 2020. Finally, we confirmed that New Zealand transmission clusters identified via standard health surveillance strategies largely agree with those defined by genomic data. We have demonstrated how the use of genomic data and computational biology methods can assist health officials in characterising the epidemiology of viral epidemics and for contact tracing.

Locally Optimal Subsampling Strategies for Full Matrix Capture Measurements in Pipe Inspection

In ultrasonic non-destructive testing, array and matrix transducers are being employed for applications that require in-field steerability or which benefit from a higher number of insonification angles. Having many transmit channels, on the other hand, increases the measurement time and renders the use of array transducers unfeasible for many applications. In the literature, methods for reducing the number of required channels compared to the full matrix capture scheme have been proposed. Conventionally, these are based on choosing the aperture that is as wide as possible. In this publication, we investigate a scenario from the field of pipe inspection, where cracks have to be detected in specific areas near the weld. Consequently, the width of the aperture has to be chosen according to the region of interest at hand. On the basis of ray-tracing simulations which incorporate a model of the transducer directivity and beam spread at the interface, we derive application specific measures of the energy distribution over the array configuration for given regions of interest. These are used to determine feasible subsampling schemes. For the given scenario, the validity/quality of the derived subsampling schemes are compared on the basis of reconstructions using the conventional total focusing method as well as sparsity driven-reconstructions using the Fast Iterative Shrinkage-Thresholding Algorithm. The results can be used to effectively improve the measurement time for the given application without notable loss in defect detectability.

In vivo human whole-brain Connectom diffusion MRI dataset at 760\u2009\xb5m isotropic resolution

We present a whole-brain in vivo diffusion MRI (dMRI) dataset acquired at 760 μm isotropic resolution and sampled at 1260 q-space points across 9 two-hour sessions on a single healthy participant. The creation of this benchmark dataset is possible through the synergistic use of advanced acquisition hardware and software including the high-gradient-strength Connectom scanner, a custom-built 64-channel phased-array coil, a personalized motion-robust head stabilizer, a recently developed SNR-efficient dMRI acquisition method, and parallel imaging reconstruction with advanced ghost reduction algorithm. With its unprecedented resolution, SNR and image quality, we envision that this dataset will have a broad range of investigational, educational, and clinical applications that will advance the understanding of human brain structures and connectivity. This comprehensive dataset can also be used as a test bed for new modeling, sub-sampling strategies, denoising and processing algorithms, potentially providing a common testing platform for further development of in vivo high resolution dMRI techniques. Whole brain anatomical T1-weighted and T2-weighted images at submillimeter scale along with field maps are also made available.

Assessment of Subsampling Strategies in Microspectroscopy of Environmental Microplastic Samples

The analysis of environmental occurrence of microplastic (MP) particles has gained notable attention within the past decade. An effective risk assessment of MP litter requires elucidating sources of MP particles, their pathways of distribution and, ultimately, sinks. Therefore, sampling has to be done in high frequency, both spatially and temporally, resulting in a high number of samples to analyze. Microspectroscopy techniques, such as FTIR imaging or Raman particle measurements allow an accurate analysis of MP particles regarding their chemical classification and size. However, these methods are time-consuming, which gives motivation to establish subsampling protocols that require measuring less particles, while still obtaining reliable results. The challenge regarding the subsampling of environmental MP samples lies in the heterogeneity of MP types and the relatively low numbers of target particles. Herein, we present a comprehensive assessment of different proposed subsampling methods on a selection of real-world samples from different environmental compartments. The methods are analyzed and compared with respect to resulting MP count errors, which eventually allows giving recommendations for staying within acceptable error margins. Our results are based on measurements with Raman microspectroscopy, but are applicable to any other analysis technique. We show that the subsampling-errors are mainly due to statistical counting errors (i.e., extrapolation from low numbers) and only in edge cases additionally impacted by inhomogeneous distribution of particles on the filters. Keeping the subsampling-errors low can mainly be realized by increasing the fraction of MP particles in the samples.

Learning With Fewer Images via Image Clustering: Application to Intravascular OCT Image Segmentation.

Deep learning based methods are routinely used to segment various structures of interest in varied medical imaging modalities. Acquiring annotations for a large number of images requires a skilled analyst, and the process is both time consuming and challenging. Our approach to reduce effort is to reduce the number of images needing detailed annotation. For intravascular optical coherence tomography (IVOCT) image pullbacks, we tested 10% to 100% of training images derived from two schemes: equally-spaced image subsampling and deep-learning- based image clustering. The first strategy involves selecting images at equally spaced intervals from the volume, accounting for the high spatial correlation between neighboring images. In clustering, we used an autoencoder to create a deep feature space representation, performed k-medoids clustering, and then used the cluster medians for training. For coronary calcifications, a baseline U-net model was trained on all images from volumes of interest (VOIs) and compared with fewer images from the sub-sampling strategies. For a given sampling ratio, the clustering based method performed better or similar as compared to the equally spaced sampling approach (e.g., with 10% of images, mean F1 score for calcific class increased from 0.52 to 0.63, with equal spacing and with clustering, respectively). Additionally, for a fixed number of training images, sampling images from more VOIs performed better than otherwise. In conclusion, we recommend the clustering based approach to annotate a small fraction of images, creating a baseline model, which potentially can be improved further by annotating images selected using methods described in active learning research.

A Machine Learning Approach for the Automated Interpretation of Plasma Amino Acid Profiles.

Plasma amino acid (PAA) profiles are used in routine clinical practice for the diagnosis and monitoring of inherited disorders of amino acid metabolism, organic acidemias, and urea cycle defects. Interpretation of PAA profiles is complex and requires substantial training and expertise to perform. Given previous demonstrations of the ability of machine learning (ML) algorithms to interpret complex clinical biochemistry data, we sought to determine if ML-derived classifiers could interpret PAA profiles with high predictive performance. We collected PAA profiling data routinely performed within a clinical biochemistry laboratory (2084 profiles) and developed decision support classifiers with several ML algorithms. We tested the generalization performance of each classifier using a nested cross-validation (CV) procedure and examined the effect of various subsampling, feature selection, and ensemble learning strategies. The classifiers demonstrated excellent predictive performance, with the 3 ML algorithms tested producing comparable results. The best-performing ensemble binary classifier achieved a mean precision-recall (PR) AUC of 0.957 (95% CI 0.952, 0.962) and the best-performing ensemble multiclass classifier achieved a mean F4 score of 0.788 (0.773, 0.803). This work builds upon previous demonstrations of the utility of ML-derived decision support tools in clinical biochemistry laboratories. Our findings suggest that, pending additional validation studies, such tools could potentially be used in routine clinical practice to streamline and aid the interpretation of PAA profiles. This would be particularly useful in laboratories with limited resources and large workloads. We provide the necessary code for other laboratories to develop their own decision support tools.