Xiaotao Shen

Biological aging clocks summarize molecular changes as a single age estimate, but cannot distinguish disease-associated changes that follow healthy aging from those that deviate from it. Here we introduce scCody, a generative framework that learns age-dependent cellular population dynamics from cross-sectional single-cell cohorts using population-level optimal transport and endpoint-consistency regularization. scCody accurately reconstructed cellular populations at unseen ages, recovered age-associated transcriptional changes more effectively than existing generative dynamics models, and enabled donor-age inference without direct age-regression training. Using the learned healthy aging reference, scCody decomposed disease-associated cellular displacement into aging-aligned and orthogonal components. In Alzheimer's disease, selected neuronal populations showed substantial aging-aligned remodeling alongside shared orthogonal changes associated with established pathology signatures. In contrast, COVID-19 exhibited weaker aging alignment and more pronounced orthogonal immune remodeling. Further decomposition revealed patient-level interferon variation associated with symptom duration and increased transcriptional heterogeneity in CD16 monocytes. This framework provides an extension of biological age assessment beyond a single age estimate, distinguishing cellular changes that resemble healthy aging from those that depart from it.

Untargeted liquid chromatography-mass spectrometry (LC-MS) metabolomics generates thousands of features, but incomplete metabolite annotation and reliance on predefined pathways limit their biological interpretation. Here we present FeatureMAP, an LC-MS feature-centric computational framework that integrates multi-evidence metabolite annotation, network-guided refinement, metabolic module discovery, and structured large language model-based functional interpretation. Across ten public datasets, FeatureMAP improved MS1-based metabolite candidate prioritization while reducing annotation redundancy. In semi-synthetic benchmarks, FeatureMAP recovered coordinated metabolic perturbations, achieving 92% recovery of a simulated perturbation, and distinguished concurrent metabolic processes. Controlled benchmarks further evaluated functional interpretation under competing metabolite-level signals. Based on FeatureMAP, we additionally developed PhenoMAP, a phenotype-aware knowledge graph comprising 12,929 dysregulated metabolic modules from 501 human metabolomics studies spanning 326 phenotypes. Module-level comparisons revealed reproducible metabolic similarities across independent studies and biologically supported relationships among phenotypes. Application to Alzheimer’s disease cerebrospinal fluid and cellular datasets identified convergent membrane-lipid alterations and implicated galactose-linked glycosphingolipid remodeling. Together, FeatureMAP and PhenoMAP enable network-grounded interpretation of incompletely annotated LC-MS metabolomics data and systematic investigation of metabolic dysregulation across biological conditions.

Multi‐omics technologies generate high‐dimensional molecular signatures that provide unprecedented opportunities to uncover biological mechanisms. However, translating complex molecular alterations into coherent and interpretable functional insights remains a major challenge. Existing module discovery methods can identify groups of related features, but often lack direct biological interpretability, whereas pathway‐based approaches frequently yield redundant results that complicate interpretation. Here, we present MAPA (Modular Analysis and Phenotype‐informed Annotation using large language models [LLMs]), a semantic‐biological network framework for functional module discovery and interpretation in multi‐omics data. MAPA integrates molecular interactions and pathway‐level functional context into a unified semantic‐biological network, and applies random walk with restart to quantify global functional relatedness among molecules and pathways for coherent module discovery across omics layers. MAPA further incorporates LLM‐assisted interpretation with retrieval‐augmented generation (RAG) to produce structured, literature‐informed module interpretation. Benchmarking against existing approaches shows that MAPA achieves superior module reconstruction and expert‐aligned functional interpretation. Applied to aging‐related multi‐omics datasets, MAPA reveals biologically coherent modules and biological insights that are difficult to obtain from conventional pathway analyses alone. MAPA provides a generalizable framework for organizing fragmented and heterogeneous molecular features into functional modules and comprehensive interpretations.

Inferring causal relationships from time-series data is critical for uncovering the dynamics of biological regulation. However, in multi-omics studies, this task is often hampered by sparse temporal sampling and the limitations of existing methods. To address this, we developed Lagged-Correlation Based Causal Inference (lagCI), a computational framework designed to identify time-lagged associations by combining comprehensive lag-correlation profiling with a robust statistical filtering scheme. Rather than relying on simple cross-correlation, lagCI analyzes the entire correlation profile and applies a quality-scoring system to filter out spurious associations that often plague high-dimensional datasets. We first tested lagCI on wearable physiological data, where it successfully captured the well-known causal link between physical activity and heart rate, even accounting for variations in lag times between individuals. Moving to high-frequency human multi-omics, we used lagCI to build a directed network of 1624 molecules connected by over 157 000 predicted interactions. This network recapitulated established biological relationships, including cytokine–hormone crosstalk, and highlighted molecular hubs that may coordinate the timing of metabolic and immune responses. Overall, lagCI provides a data-driven way to extract temporal insights from dense longitudinal omics. The tool is available as an R package with multiple interfaces to support use by both bioinformaticians and clinical researchers.

A growing body of literature has investigated the relationship between age and biomolecular changes, leading to conclusions that aging occurs in discrete molecular ``waves.'' Data summary tools such as LOESS and sliding window analyses like DE-SWAN are common approaches that have gained acceptance in recent years. We demonstrate via simple simulations that these tools can identify non-linear patterns of aging where they do not exist. Specifically, we show that (i) clustering of molecular trajectories using LOESS can lead to artifactual characteristic patterns of molecular aging, (ii) ``waves'' of aging identified using the combination of LOESS and DE-SWAN in real data are not robust to changes in the underlying age distribution and are not supported by valid permutation testing, and (iii) DE-SWAN alone can generate pronounced "waves'" of nonlinear molecular aging in linear data due to differences in statistical power along the age continuum. Our results specifically challenge the statistical support for discrete aging crests inferred in the literature, but do not rule out nonlinear molecular aging or age-associated transitions that may be detectable using other cohorts and statistical models.

Liquid chromatography-mass spectrometry (LC-MS) untargeted metabolomics detects thousands of metabolic features, but converting these chemical signals into metabolite set-level biological knowledge remains challenging. This is because most features lack unambiguous metabolite identities. Conventional metabolite set enrichment analysis (MSEA) generally requires identified metabolites and metabolite-level ranked inputs, leaving much of the untargeted feature space unused. Here, we present FeatureMSEA, a feature rank-based framework for metabolite set enrichment directly from metabolic features with ambiguous annotations. FeatureMSEA integrates multi-evidence feature-to-metabolite annotation, feature rank-based enrichment scoring, permutation-based inference, and iterative leading-edge-guided annotation refinement, with an optional LLM-assisted module for post-enrichment interpretation. In null comparisons of randomly split healthy samples, FeatureMSEA detected no significant metabolite sets, whereas metabolite-set spike-in simulations showed recovery of implanted signals. In a cerebrospinal fluid metabolomics study of Huntington's disease, FeatureMSEA identified dysregulated metabolite sets related to amino acid metabolism, mitochondrial energy metabolism, and neuroactive signaling. MS/MS-based annotation analysis further showed that FeatureMSEA refinement reduced annotation ambiguity and prioritized chemically consistent candidate metabolites. In summary, FeatureMSEA provides a general framework for extracting metabolite set-level biological insights from LC-MS untargeted metabolomics in which confident metabolite identification remains incomplete.

Inferring causal relationships from time-series data is critical for uncovering the dynamics of biological regulation. However, in multi-omics studies, this task is often hampered by sparse temporal sampling and the limitations of existing methods. To address this, we developed Lagged-Correlation Based Causal Inference (lagCI), a computational framework designed to identify time-lagged associations by combining comprehensive lag-correlation profiling with a robust statistical filtering scheme. Rather than relying on simple cross-correlation, lagCI analyzes the entire correlation profile and applies a quality-scoring system to filter out spurious associations that often plague high-dimensional datasets. We first tested lagCI on wearable physiological data, where it successfully captured the well-known causal link between physical activity and heart rate, even accounting for variations in lag times between individuals. Moving to high-frequency human multi-omics, we used lagCI to build a directed network of 1,624 molecules connected by over 157,000 predicted interactions. This network didn’t just mirror established biology (such as cytokine-hormone crosstalk); it also pointed to specific molecular hubs that seem to orchestrate the timing of metabolic and immune responses. Overall, lagCI provides a data-driven way to extract temporal insights from dense longitudinal omics. We’ve made the tool available as an R package with multiple interfaces to ensure it’s accessible for both bioinformaticians and clinicians.

The human microbiome profoundly influences the host plasma metabolome and health, but most studies have focused on the gut microbiome in isolation. A compr

Aims/hypothesis The aim of this study was to identify conserved biomarkers for diabetic retinopathy progression, addressing the lack of minimally invasive biomarkers to track the entire continuum of diabetic retinopathy. Methods We conducted high-throughput proteomics (SomaScan v4.1) on aqueous humour from the Guangdong DR Multiple-Omics Study (n=32). Temporal protein clusters were identified using soft clustering of temporal trajectories. Candidates were validated in a US dataset and localised via single-cell RNA-seq in an oxygen-induced retinopathy mouse model. The clinical relevance of plasma-detectable markers was assessed cross-sectionally and prospectively in individuals with diabetes from the UK Biobank (n=2495). Results In the discovery dataset (mean age 68.5±9.0 years; 56.3% men, 43.8% women), temporal proteomic profiling identified 40 candidate biomarkers whose concentration showed monotonic changes during diabetic retinopathy progression. Of these, 25 showed conserved directional patterns in the validation dataset (17 increasing and eight decreasing, all p<0.05). Single-cell mapping further localised 15 candidates, including neurofilament light chain (NFL), to retinal neurons and glia. In the UK Biobank individuals, baseline plasma NFL distinguished those with diabetic retinopathy from control individuals (OR 1.98 [95% CI 1.61, 2.42]) and predicted incident diabetic retinopathy (HR 2.01 [95% CI 1.48, 2.73]) and vascular complications (microvascular, HR 2.28 [95% CI 1.94, 2.69]; macrovascular, HR 1.49 [95% CI 1.26, 1.77]) over a median follow-up of 12 years. Plasma NFL enhanced the predictability of a conventional risk factor model for diabetic retinopathy (net reclassification improvement [NRI] 0.194 [95% CI 0.042, 0.297]; integrated discrimination improvement [IDI] 0.015 [95% CI 0.003, 0.047]) and enabled risk stratification of vascular complications. Conclusions/interpretation NFL represents a pan-stage biomarker for diabetic retinopathy progression, detectable through minimally invasive plasma testing, and is associated with risks of diabetic vascular complications. Graphical Abstract