The research, published on 16 October 2026, draws on publicly available data from the Single-cell Pediatric Cancer Atlas, or ScPCA, a resource that catalogs single-cell profiles from pediatric cancer patients. Kim reprocessed two public leukemia projects from the atlas using a common cell-type mapping strategy, ensuring that cells from different studies could be compared on the same biological vocabulary. From this harmonized foundation, a cross-project immune composition was constructed from three major compartments: T and natural killer cells, B-lineage cells, and myeloid cells. The goal was to see whether these proportions, tracked over time within individual patients, could reveal something about the trajectory of the disease that simpler categorical labels would miss.

Central to the study is the concept of ecological context. In cancer biology, the tumor microenvironment is often described in terms of discrete states, such as immune-hot versus immune-cold, or dominant cell type versus suppressed lineage. These coarse labels are convenient, but they risk discarding information. Kim’s framework treats immune composition as a continuous variable embedded in a stochastic longitudinal model, specifically an Ornstein-Uhlenbeck process, which is a mathematical model that describes a value fluctuating around a long-term mean with a tendency to revert toward that mean. In this context, the OU process models the leukemia-state trajectory over time, and the question is whether immune composition shifts the attractor, the value mu toward which the trajectory is pulled.

Among 104 compositionally eligible participants, coarse clustering with two groups was reproducible under bootstrap resampling, meaning that the split into two clusters was stable when the data were resampled repeatedly. However, this clustering showed only limited association with the project from which the data came or with the specific diagnosis. Within the 99 participants with B-cell acute lymphoblastic leukemia, the most common pediatric leukemia, the two clusters separated 73 participants with T and natural killer cell dominant profiles from 26 participants with B and myeloid shifted profiles. This separation was largely independent of which project the data originated from, suggesting a genuine biological structure rather than a batch effect.

Yet a critical complication emerged. When Kim compared cluster assignments derived from raw composition with those derived from centered log-ratio transformed composition, a standard transformation for compositional data that accounts for the constraint that proportions must sum to one, the concordance was poor. This means that the apparent ecological state a patient belongs to depends heavily on how the compositional data are represented mathematically. A patient might appear in one cluster under raw proportions and in a different cluster under log-ratio transformed proportions. This representation sensitivity undermines the reliability of coarse context labels as stable descriptors of biological reality.

Because of this instability, Kim chose to evaluate continuous ecological composition separately from coarse context labels in a corrected longitudinal analysis. The longitudinal dataset included 34 participants contributing 241 observations and 207 transitions between time points. The key finding was that higher B-lineage composition was associated with a lower effective OU attractor, the parameter mu, in both models restricted to B-lineage cells alone and models that included B plus myeloid compartments. In biological terms, this suggests that as B-lineage cells become more prevalent in the immune environment, the leukemia state trajectory is pulled toward a lower value, potentially reflecting a less aggressive or more controlled disease state.

In contrast, the coarse context effects, the discrete labels that would traditionally be used to categorize patients, were unresolved in the longitudinal models. They did not provide clear, consistent information about the direction or magnitude of the leukemia trajectory. This is a striking result because it suggests that the very labels clinicians and researchers often rely on to summarize immune context may be less informative than the underlying continuous composition they are meant to represent.

To assess whether the continuous composition genuinely adds predictive value, Kim employed leave-one-participant-out validation, a rigorous approach in which each participant is held out in turn and the model is trained on the remaining participants. In aggregate, the model including B-lineage composition modestly outperformed the OU-only model that lacked ecological covariates. However, the predictive differences were heterogeneous across participants, meaning that for some individuals the continuous composition helped substantially while for others it did not. Models that included only coarse context labels, or that combined ecology with context, provided no incremental advantage over the simpler approaches. This pattern reinforces the conclusion that the value lies in the continuous signal, not in the categorical summaries.

A secondary analysis using an ordinal-state model, which treats the leukemia state as an ordered categorical variable rather than a continuous one, revealed a directionally consistent B-associated signal. This consistency across different modeling frameworks strengthens confidence that the B-lineage association is not an artifact of a single analytical choice. Simulations using the observed irregular sampling schedules from the real data demonstrated that moderate ecological effects could be recovered even when observations were not evenly spaced in time, addressing a common concern in longitudinal single-cell studies. Posterior-predictive analyses further showed that including continuous ecological composition improved the calibration of uncertainty estimates, meaning that the model’s confidence intervals better reflected actual variability in the data.

The broader implications of this work extend beyond pediatric leukemia. The study demonstrates a general principle: stable clustering of compositional data does not necessarily imply that those clusters represent longitudinally predictive ecological states. Researchers across cancer biology, immunology, and microbial ecology often face the temptation to reduce continuous compositional information into discrete categories for ease of interpretation. Kim’s framework offers a rigorous way to test whether such reduction loses information, and it provides a template for distinguishing descriptive ecological coarse-graining from ecological covariates that genuinely add explanatory power to stochastic longitudinal models. Importantly, the author cautions that the OU mean reversion observed in the models should not be interpreted as direct evidence of stabilizing selection or of fixed leukemia ecotypes, a nuance that guards against overinterpretation of the mathematical framework. As single-cell technologies continue to generate rich longitudinal datasets, approaches like this one will be essential for extracting the full biological signal from the immune landscapes that surround cancer.

Subject of Research: Continuous immune composition as a longitudinal predictor in pediatric leukemia single-cell data

Article Title: Continuous immune composition retains longitudinal information beyond coarse ecological-context labels in pediatric leukemia

Article References: Kim, S.-H. (2026). Continuous immune composition retains longitudinal information beyond coarse ecological-context labels in pediatric leukemia. BMC Bioinformatics. https://doi.org/10.1186/s12859-026-06679-7

Image Credits: AI Generated

DOI: 10.1186/s12859-026-06679-7

Keywords: pediatric leukemia, single-cell RNA-seq, immune composition, Ornstein-Uhlenbeck process, Bayesian inference, tumor microenvironment, B-cell acute lymphoblastic leukemia, longitudinal modeling, compositional data, ScPCA, ecological modeling, BMC Bioinformatics

The research, published in Discover Geoscience by Dionisio Uendro Carlos, César Augusto Moreira, and Guilherme Libardi Maestrelo, introduces a moving-window ordinary kriging (MW-OK) algorithm designed specifically for airborne and ground radiometric data. Conventional interpolators such as inverse distance weighting and minimum curvature are fast and widely used, but they share a fundamental weakness: they do not explicitly account for the local spatial structure of the data. In practice, that means they can oversmooth short-wavelength variability, blurring exactly the sharp radiometric gradients that mark boundaries between rock units or zones of hydrothermal alteration.

Kriging, by contrast, is a geostatistical method built on a statistical model of how a property varies in space. Because radioactive decay is inherently a stochastic process, the authors argue, an interpolator that respects this randomness is desirable. Yet standard kriging typically relies on a single global variogram, the mathematical function describing spatial correlation, assumed to hold across the entire survey area. Radiometric variables rarely obey that assumption at regional scale. Lithological boundaries, weathering domains, and survey-related effects all conspire to make the spatial structure change from place to place, so a global model can over-smooth local features or produce biased estimates where the structure shifts.

MW-OK sidesteps this problem by abandoning the global model altogether. For every prediction node on the output grid, the algorithm defines a circular window of investigation, in this case with a radius of 750 meters, roughly 1.5 times the 500-meter flight-line spacing, so that each window intersects two adjacent flight lines. Within that window, the method assumes second-order stationarity holds only approximately and locally. An empirical semivariogram is computed from all unique pairs of samples in the neighborhood, binned into eight lag classes, with any bin containing fewer than three valid pairs rejected. An exponential model is then fitted to each local semivariogram by bounded non-linear least squares, yielding a nugget effect, partial sill, and range parameter tailored to that specific patch of ground.

The computational scale is striking. The study area, a 40 by 40 kilometer subset of the Sul-Riograndense Shield Aerogeophysical Project flown by the Geological Survey of Brazil in 2010, contains 50,374 measurement points. Because the algorithm estimates a local variogram for each measurement point, it effectively generates 50,374 variograms. Neighbor retrieval is accelerated with a k-dimensional tree using ball queries, and safeguards keep the local kriging systems well conditioned: a minimum of eight neighbors, a maximum of thirty, removal of duplicate coordinates to prevent singular matrices, and a tiny ridge term added to the kriging matrix diagonal for numerical robustness. If a non-linear variogram fit fails, the workflow falls back on physically sensible defaults derived from the local data variance and median nearest-neighbor spacing.

The fitted local variograms themselves carry geological information. For potassium, the nugget effect was essentially zero at the scale of the data support, meaning microscale variability and measurement noise are minor compared with the spatially structured component, and correlation decayed over a practical range of about 1.3 kilometers. Equivalent uranium showed an even longer continuity, around 1.7 kilometers, consistent with smoother spatial behavior, while equivalent thorium displayed a larger partial sill, indicating higher structured variance, with a moderate range near 1.5 kilometers. These parameters confirm that, within each window, the radiometric variability is predominantly spatially structured rather than random noise.

When the researchers applied MW-OK to the survey data, the resulting potassium, uranium, and thorium maps reproduced the broad patterns that geologists expect. Potassium enrichment above 4 percent traced the São Sepé Granite in the north, elevated values of roughly 4 percent marked the Ramada Granite in the southwest, and low concentrations coincided with the metatuff-dominated Bossoroca Complex and the sandstones of the Maricá Group. Thorium highs of about 21 parts per million again outlined the São Sepé Granite, with additional anomalies over the Cerro da Cria Granite and nearby rhyolites. But the kriged surfaces went further than the conventional minimum-curvature grids: they better represented local gradient variations near lithological contacts and revealed internal heterogeneity within the felsic bodies that the smoother deterministic algorithm muted. Notably, three east-west linear artifacts aligned with the survey tie lines, visible in the minimum-curvature uranium map, were suppressed in the kriged surface.

A key advantage of kriging is that it also delivers an estimate of its own uncertainty. The kriging variance maps for all three radioelements bore the clear imprint of the acquisition geometry: uncertainty was minimized along and near the north-south flight lines and increased in the gaps between them, telling interpreters exactly where the estimates are most reliable. To quantify how closely MW-OK matched the industry-standard approach, the team performed pixel-wise comparisons with the minimum-curvature grids. The agreement was remarkably strong, with coefficients of determination of 0.9938 for potassium, 0.9727 for equivalent uranium, and 0.9946 for equivalent thorium, and residual histograms tightly centered near zero. Ordinary least-squares fits revealed a subtle but systematic amplitude scaling, with slopes slightly below unity, suggesting MW-OK modestly attenuates the highest values while remaining close to the identity overall. The authors are careful to frame this as evidence of methodological consistency rather than broad superiority: the two methods agree at the domain scale, and their differences emerge primarily at local scales.

Beyond the three radioelement grids, the workflow generates the interpretive products that exploration geologists actually use. It computes standard radiometric ratios (K/eTh, K/eU, eU/eTh) and the F-parameter, a widely used proxy for hydrothermal alteration that relates potassium abundance to the uranium-to-thorium ratio. It also produces ternary radioelement maps in three complementary forms: the classical K-eU-eTh display, normalized ternaries, and compositional-data transformations using additive, centered, and isometric log-ratios that respect the closure constraint and reduce spurious correlation when treating the three elements as parts of a composition. In the study area, a historically significant gold district in southern Brazil where mineralization is hydrothermal and structurally controlled in quartz veins hosted by schists and granites, these products delineated the main lithological contacts and showed spatial correspondence with known gold-mineralized zones and areas of gold enrichment identified by stream-sediment prospecting.

Perhaps most importantly for the wider community, the entire workflow is open source and exports standard deliverables, including GXF grids, GeoTIFF rasters, KMZ overlays, and PNG figures, with proper spatial metadata, making it directly interoperable with both open-source tools such as Python, QGIS, and Google Earth and commercial platforms like Oasis Montaj, Intrepid Geophysics, and ArcGIS. The authors acknowledge the method’s limitations: effective spatial resolution remains constrained by flight-line spacing and sampling interval, computational cost grows with finer grids and larger neighborhoods, and sparsely sampled sectors can yield under-populated windows with unstable variograms, which the workflow mitigates through minimum-neighbor thresholds and k-nearest-neighbor fallbacks. Looking ahead, they plan to incorporate co-regionalized cokriging constrained by lithology and structure, and deep-kriging approaches, within the same moving circular-window framework. For now, MW-OK stands as a robust, locally adaptive alternative that offers incremental rather than transformative gains over deterministic gridding, but gains that could make the difference between a blurred contact and a clearly resolved exploration target.

Subject of Research: A locally adaptive geostatistical interpolation method for processing airborne and ground gamma-ray spectrometric data

Article Title: A moving-window ordinary kriging approach for airborne and ground gamma-ray spectrometric data processing

Article References: Carlos, D. U., Moreira, C. A., & Maestrelo, G. L. (2026). A moving-window ordinary kriging approach for airborne and ground gamma-ray spectrometric data processing. Discover Geoscience, 4(1), Article 289. https://doi.org/10.1007/s44288-026-00674-4

Image Credits: AI Generated

DOI: 10.1007/s44288-026-00674-4

Keywords: gamma-ray spectrometry, kriging, geostatistics, radiometric mapping, mineral exploration, semivariogram, interpolation, gold prospecting, geological mapping, Brazil, open-source software, compositional data

The research team, led by Sunyong Park and colleagues including Sunhwa Ryu, assembled a literature-derived dataset of 552 thermally treated biomass samples, spanning torrefaction and pyrolysis experiments drawn from dozens of published studies. Each record contained ultimate analysis, proximate analysis, higher heating value, and the process conditions—treatment temperature and residence time—that produced the fuel. Rather than feeding these numbers raw into an algorithm, the researchers invested heavily in preprocessing. Because both ultimate and proximate analyses are compositional data whose variables must sum to 100 percent, they applied closure normalization separately to each block and computed centered log-ratio transforms, a technique from compositional statistics that suppresses the spurious correlations that fixed-sum data otherwise generate.

Feature engineering went further. The team built process-aware descriptors—squared temperatures, logarithmic residence times, temperature–time interaction terms, and a combined severity index—to capture the nonlinear way thermal treatment intensity drives devolatilization, oxygen removal, and carbon enrichment. Composition-aware descriptors such as H/C and O/C ratios encoded the degree of carbonization, while proximate-derived quantities like the volatile-matter-to-fixed-carbon ratio captured the balance between volatile release and char formation. Critically, the authors also built in safeguards against information leakage and circular reasoning: auxiliary features were constructed only from training data, and a Dulong-equation-derived heating value surrogate was deliberately excluded from the final heating value models to ensure the algorithm was not simply copying a textbook formula hidden inside its own inputs.

After outlier removal using the interquartile range rule—which cut 234 records, or 42.4 percent of the dataset, leaving 318 cleaned samples—the team trained and compared linear, bagging, boosting, and tree-ensemble models, including Ridge regression, partial least squares, Random Forest, Extra Trees, XGBoost, and CatBoost. Each target variable got its own single-output model, selected by validation performance and tuned with grid search. To avoid the common trap of reporting a flattering single data split, the researchers re-evaluated every selected model across five repeated random train-validation-test partitions and reported the mean and standard deviation of test-set performance.

The results reveal a striking asymmetry between the two prediction directions. Predicting proximate properties and heating value from elemental composition—the ultimate-to-proximate direction—worked well. CatBoost models achieved repeated test-set R-squared values of 0.756 for volatile matter and 0.751 for higher heating value, with XGBoost reaching 0.702 for fixed carbon and Extra Trees 0.607 for ash. In plain terms, knowing what elements a treated biomass contains, along with how hot and how long it was treated, is enough to estimate its volatile content and energy density with useful accuracy. The heating value model predicted values with a mean absolute error of roughly 0.75 megajoules per kilogram—impressive for a screening tool that never lights a flame.

Going the other way proved harder. When the models tried to infer elemental composition from proximate data, performance dropped and became far more sensitive to how the data were partitioned. Carbon, hydrogen, oxygen, and heating value all landed in a moderate range, with repeated R-squared values between roughly 0.41 and 0.57, and large standard deviations betraying instability across splits. The reason is mathematical as much as chemical: proximate analysis compresses the fuel into three broad fractions, and many different elemental combinations can produce nearly identical volatile, fixed-carbon, and ash profiles. The inverse problem is simply underdetermined. Sulphur fared worst of all, with a negative mean R-squared of −0.085, meaning the model performed worse than naively guessing the dataset average. The authors are refreshingly blunt about this: sulphur, present at very low and heavily skewed concentrations, is flagged as unreliable and excluded from quantitative decision-making, and nitrogen is marked as a low-confidence screening estimate only.

The team also stress-tested their own pipeline. A filtering sensitivity analysis compared no outlier removal, gentle filtering, and the strict final setting, and found the effects were target-dependent rather than uniformly beneficial. Volatile matter and heating value barely changed across scenarios, but ash prediction improved substantially with stricter filtering—its R-squared rising from 0.44 to 0.64—because extreme ash values from heterogeneous feedstocks were dragging the model off course. Meanwhile, several inverse-direction targets actually got worse under strict filtering, suggesting that aggressive pruning had removed compositional variability the models needed. This nuanced outcome matters: it demonstrates the reported performance was not an artifact of quietly discarding nearly half the data, while honestly acknowledging that the final models cover a narrower slice of the biomass universe than the raw literature does.

The benchmark against human-devised formulas may be the study’s most compelling result. On the same held-out test set, the Park et al. empirical equations for predicting proximate properties from elements managed R-squared values of 0.72 for volatile matter and 0.40 for ash, while the machine-learning models reached 0.89 and 0.64 respectively. The gap widened dramatically for heating value: the century-old Dulong equation produced a negative R-squared of −1.01 on this thermally treated dataset, whereas the Dulong-free machine-learning model scored 0.87. In the inverse direction, the Shen et al. correlations from 2010 essentially failed on modern torrefied biomass, with carbon prediction collapsing to an R-squared of −0.54, while the machine-learning framework achieved 0.62. The lesson is that fixed-form equations derived decades ago on different feedstocks simply do not transfer to the complex, treatment-dependent fuels now entering the market, whereas flexible nonlinear models trained on relevant data do.

Interpreting the models with SHAP analysis added physical plausibility to the statistics. For heating value prediction, carbon- and oxygen-related descriptors—including closure-aware carbon terms and O/C variables—dominated the feature importance, exactly matching the established chemistry that carbon enrichment and deoxygenation raise energy density during torrefaction. Severity-weighted terms and temperature–residence-time interactions also contributed, indicating the models had genuinely learned treatment-intensity effects rather than memorizing feedstock labels. For ash, the influential features were hydrogen, nitrogen, and sulphur descriptors, which the authors interpret not as causal drivers of ash but as statistical fingerprints of feedstock type and inorganic-rich sample groups—a candid reminder that SHAP values describe model behavior, not experimental mechanism.

To make the framework usable beyond the modeling team, the researchers wrapped it in a lightweight Streamlit graphical interface that accepts either ultimate or proximate inputs plus process conditions and returns cross-predictions with heating value estimates, input validation, compositional closure checks, and warnings when users venture outside the training domain. Reliability flags are baked in: sulphur outputs are explicitly marked unusable for quantitative decisions. The authors are careful to position the tool as a screening aid, not a laboratory replacement—regulatory evaluation of low-concentration elements still demands real analysis, and the literature-derived dataset carries inherent heterogeneity in feedstock, reactor design, and analytical protocol. But for traders, plant operators, and researchers needing a fast preliminary read on whether a biomass batch is worth pursuing, the message is clear: a well-engineered machine-learning model, honest about its own uncertainty, can now deliver in seconds what once required a destructive, instrument-intensive afternoon—and it can do it more accurately than the empirical formulas the field has relied on for decades.

Subject of Research: Machine-learning cross-prediction of ultimate and proximate properties of thermally treated biomass for cleaner fuel assessment

Article Title: Machine-Learning-Based Cross-Prediction of Ultimate and Proximate Properties for Cleaner Biomass Fuel Assessment

Article References: Park, S., Lee, J., Yang, J., Kim, S., Hwang, K., & Ryu, S. (2026). Machine-Learning-Based Cross-Prediction of Ultimate and Proximate Properties for Cleaner Biomass Fuel Assessment. Cleaner Engineering and Technology, Article 101333. https://doi.org/10.1016/j.clet.2026.101333

Image Credits: AI Generated

DOI: 10.1016/j.clet.2026.101333

Keywords: biomass, machine learning, torrefaction, proximate analysis, ultimate analysis, higher heating value, fuel quality, compositional data, SHAP, XGBoost, CatBoost, cleaner energy

The challenge that Joint-RPCA tackles is deceptively simple to state and notoriously hard to solve. Microbiome data are compositional, meaning sequencing reflects relative rather than absolute abundances. They are sparse, with many measurements containing zeros or missing values. And they span wildly different scales, because metabolomics, proteomics, and genomics generate numbers in different units and magnitudes. Traditional approaches often sidestep these problems by analyzing each data layer separately, using dimensionality reduction techniques such as principal coordinates analysis on a distance matrix. But doing so treats the data layers as independent and ignores the inherent correlations between modalities sampled from the same ecosystem, such as the production of a specific metabolite by a specific bacterium.

Joint-RPCA, developed by Bianca Cordazzo Vargas, Cameron Martino, Liat Shenhav, and colleagues spanning institutions from New York University to the University of California San Diego, the University of Turku, and Ben-Gurion University, addresses these issues head-on. The method builds on the OptSpace matrix completion framework, assuming that the input data matrices share an underlying low-rank structured component. In practice, this means the dominant biological signal, such as the difference between diseased and healthy individuals, can be captured by a small number of latent factors, even when that signal is embedded in a sea of high-rank biological and technical noise. The authors are careful to clarify that low-dimensional signal refers to the dimensionality of the latent phenotype factor, not the fraction of total variance it explains, a distinction that matters greatly in microbiome settings where disease effects may be subtle relative to interpersonal variation.

Mathematically, the workflow begins by transforming each data modality using a robust centered log-ratio transformation, which handles sparsity and compositionality without requiring imputation or pseudocounts. Then, a joint dimensionality reduction is performed via singular value decomposition optimized on a local manifold. The key architectural choice is that the sample space is estimated jointly across all modalities, while the feature space is estimated individually within each modality. This yields a shared scores matrix for subjects and distinct loadings matrices for features in each omic type. The output includes a joint low-dimensional representation of samples, feature loadings indicating each feature’s contribution to the axes of variation, and a denoised feature-feature covariance matrix that can be interpreted as a multipartite network of cross-modal interactions.

To evaluate the method, the team benchmarked it against widely used microbiome approaches, including PCoA with Bray-Curtis and Aitchison distances and the single-modality RPCA, as well as general-purpose multi-omics tools such as MOFA+, iClusterPlus, intNMF, and multiblock sPLS from mixOmics. Using data-driven simulations anchored in real data from the Integrative Human Microbiome Project, with induced sparsity ranging from 12 percent down to 3 percent observed density, Joint-RPCA consistently recovered inflammatory bowel disease-associated structure more reliably than the alternatives. Classification accuracy improved by up to sixfold in this benchmark setting, and the method showed greater Mahalanobis distances between phenotype centroids and higher PERMANOVA pseudo-F statistics across simulated densities.

Consistency of feature selection proved to be another strength. When the researchers compared the top-ranked metabolomic features identified by Joint-RPCA and MOFA+ along the component most strongly associated with IBD diagnosis, Joint-RPCA showed a 60 to 70 percent median overlap in its selections across train-test splits and sparsity levels, whereas MOFA+ exhibited less than 30 percent overlap. This indicates that Joint-RPCA more stably recovers the shared low-dimensional phenotype-associated signal rather than selecting features that fluctuate with each training set. In purely synthetic simulations with traceable signals, Joint-RPCA reliably ranked the specific features carrying induced signals among its top loadings, while competitors showed variable or inconsistent performance, particularly when signals were weak.

The method also excels at recovering biologically validated cross-modal relationships. In a study of biological soil crusts, thin living layers on arid soil surfaces, roughly 70 percent of the microbe-metabolite relationships following a wetting event had been experimentally validated, providing ground truth for benchmarking. Joint-RPCA correctly assigned positive covariance values between the cyanobacterium Microcoleus vaginatus and all metabolites the isolate was known to release, and these metabolites ranked among the top 40 co-varying molecules out of 85 total. This finding remained robust even when the sequencing data were subsampled from 50 percent dense down to 1 percent dense. Compared with correlation-based approaches and the specialized method MMvec, Joint-RPCA and MMvec both achieved significantly higher true-positive rates, precision, and recall, but Joint-RPCA did so more than 100 times faster.

That speed advantage stems from a fundamental design difference. MMvec estimates conditional probabilities between metabolite abundances and microbial reads on a per-read basis, so its runtime scales linearly with the number of sequencing reads. Joint-RPCA resolves the high-dimensionality challenge by construction and operates on a per-sample basis, so its runtime scales with the number of samples instead. In runtime experiments using the FINRISK study, one of the largest multi-omics microbiome cohorts to date with 7,167 individuals, Joint-RPCA completed analyses in minutes that would take MMvec days. When processing two independent IBD cohorts with multiple omic types, MMvec required pairwise analysis of all omic combinations, and some pairs had to be excluded because runtimes exceeded 24 hours, forcing the researchers to extrapolate. Joint-RPCA processed thousands of samples and features across omic types within seconds to minutes.

Applied to real-world data, the method delivered replicable biological findings. In the iHMP dataset, which includes matched metabolomics, proteomics, viromics, metagenomics, and metatranscriptomics from 135 samples, Joint-RPCA separated IBD from non-IBD subjects across all omic layers with a PERMANOVA pseudo-F of 17.04. It identified cross-modal markers including urobilin metabolites and Klebsiella pneumoniae that aligned with prior iHMP findings, and it succeeded where single-modality RPCA failed, using the context of proteomics, viromics, and metagenomics to reveal disease-associated patterns in metatranscriptomic and metabolomic data that were otherwise invisible. The signal replicated in an independent UCSD cohort of 146 subjects, with a significant correlation of 0.47 between the IBD-associated metagenomic feature rankings of the two cohorts, and it held in a third validation dataset combining IBD patients with 824 controls from the American Gut Project. Across all three datasets, Phocaeicola vulgatus emerged as the top bacterial species associated with IBD, consistent with its experimentally demonstrated production of disease-linked proteases.

The method’s reach extends beyond human disease. Applied to multi-omics data from human cadaver decomposition across three forensic facilities, Joint-RPCA captured the progression of accumulated degree days along its second principal component and identified a universal microbial decomposer network shared across geographically and climatically distinct sites, including fungal taxa such as Yarrowia and Candida and the bacterium Thiopseudomonas alkaliphila, which single-modality analyses missed. In mammalian gut microbiomes spanning 25 species and five omic types, Joint-RPCA amplified weak signals from gas chromatography-mass spectrometry metabolomics, improving classification of host taxonomy and digestive strategy from an AUC-ROC of 0.65 with independently ranked features to 0.88 when features were ranked in the joint context. The authors caution that the method assumes a shared low-dimensional structure across modalities, does not explicitly adjust for confounders, and does not directly model temporal dynamics, and they emphasize that their benchmark results should be read as evidence of performance in the specific scenarios studied rather than a universal ranking of integration tools. Still, with open-source implementations in Python through the gemelli package and in R through the mia Bioconductor package, plus a QIIME2 plugin and Galaxy integration, Joint-RPCA arrives as a practical, scalable tool poised to reshape how microbiome scientists read the interconnected layers of microbial ecosystems.

Subject of Research: A domain-aware multi-omics integration method for systems microbiology that jointly factorizes sparse, compositional microbiome data layers.

Article Title: Joint-RPCA: domain-aware multi-omics integration for systems microbiology

Article References: Cordazzo Vargas, B., Martino, C., Dilmore, A. H., Metcalf, J. L., Burcham, Z. M., Lahti, L., Bektanov, A., Borman, T., Salomaa, V., Niiranen, T., Havulinna, A. S., Gregor, R., Eyal, S., Meijler, M. M., Mizrahi, I., Song, S. J., Bartko, A., Dorrestein, P. C., Morton, J. T., … Shenhav, L. (2026). Joint-RPCA: domain-aware multi-omics integration for systems microbiology. Molecular Systems Biology. https://doi.org/10.1038/s44320-026-00236-3

Image Credits: AI Generated

DOI: 10.1038/s44320-026-00236-3

Keywords: multi-omics integration, microbiome, Joint-RPCA, dimensionality reduction, matrix completion, inflammatory bowel disease, metabolomics, metagenomics, compositional data, microbe-metabolite interactions, systems microbiology, computational biology