Biology subjects

Kang, M.

Publications and source records attributed to Kang, M..

At least 19 recordsLinked to original sources

bioRxiv2026

Imbalance between RNA-binding proteins and the WAVE regulatory complex underlies the toxic gain-of-function effects of the CYFIP2 Arg87Cys variant in neurodevelopmental disorders

Pathogenic variants in cytoplasmic FMR1-interacting protein 2 (CYFIP2) cause neurodevelopmental disorders, with hotspot substitutions at arginine 87 (Arg87) associated with particularly severe symptoms. CYFIP2 functions both as a component of the WAVE (Wiskott-Aldrich syndrome protein family verprolin-homologous protein) regulatory complex (WRC), regulating the actin cytoskeleton, and as an interactor of RNA-binding proteins (RBPs) involved in mRNA processing/translation. Although Arg87 variants reduce CYFIP2 protein stability, heterozygous knock-in mice carrying Arg87Cys (Cyfip2+/R87C) show markedly more severe phenotypes than heterozygous knockout (Cyfip2+/-) mice, supporting a toxic gain-of-function mechanism. However, how the dual roles of CYFIP2 contribute to this mechanism remains unclear. To address this, we examined Cyfip2+/R87C mice at postnatal week 7, before overt neuronal loss and gliosis, focusing on the progressive dispersion of hippocampal cornu ammonis 1 (CA1) pyramidal neurons, a phenotype absent in Cyfip2+/- mice. Single-nucleus RNA sequencing suggested candidate alterations related to translation and cell adhesion/migration in CA1 pyramidal neurons. Complementarily, human embryonic kidney 293T cells expressing CYFIP2 Arg87Cys showed defects in proliferation and morphology, accompanied by altered expression of cell adhesion-related genes. Quantitative proteomic analysis of the CA1 region uncovered an imbalance between CYFIP2-associated protein networks, characterized by increased RBPs and reduced WRC components. Phosphoproteomic analysis further revealed altered phosphorylation of synaptic and translational regulators. Consistently, whole-cell recordings and biocytin-based morphological analyses showed increased frequency of miniature excitatory postsynaptic currents and altered dendritic morphology in Cyfip2+/R87C CA1 pyramidal neurons, most prominently in the more severely dispersed deep-layer neurons. Moreover, the cell adhesion molecule Eph receptor A3 was reduced, possibly as a result of RBP dysregulation. Together, these findings suggest that an imbalance between RBP- and WRC-associated pathways underlies the gain-of-function effects of CYFIP2 Arg87Cys and contributes to early neuronal phenotypes, most notably CA1 layer dispersion. This mechanism, distinct from CYFIP2 haploinsufficiency, supports allele-specific therapeutic strategies for Arg87 variants.

neuroscience↗

bioRxiv2026

Synthesis-aware generative design in trillion-scale chemical spaces for automated drug discovery

Autonomous drug discovery via generative molecular design is critically bottlenecked by the production of chemically intractable structures. Multi-trillion-scale make-on-demand libraries guarantee synthetic feasibility, but conventional virtual screening cannot efficiently navigate these vast spaces or address multi-parameter optimization. We introduce Hyper Screening X, a generative AI framework integrating structure-based design with automated synthesis hardware. By encoding deterministic reaction logic in a generative flow network, the framework evaluates only 10 million candidates to navigate an 11-trillion-compound space, jointly optimizing physicochemical properties while ensuring compatibility with automated synthesis. Prospective validation against an SLC1A5 variant with a cryptic interface yielded a 96% synthesis success rate and a 50% functional hit rate, identifying two first-in-class lead compounds. This synthesis-aware strategy advances autonomous, closed-loop drug discovery.

bioinformatics↗

bioRxiv2026

Cross-species spatial profiling links progranulin to an immune suppressive niche in brain metastasis

Immune checkpoint inhibitors (ICI) have revolutionized the treatment of brain metastasis (BM), demonstrating intracranial response rates and durability not previously seen with other therapies. However, approximately 50% of patients do not respond to ICI, and the brain-specific interactions that shape anti-tumor immunity remain poorly understood. We profiled five syngeneic BM models using spatial transcriptomics, identifying myeloid-rich niches enriched for interferon-responsive and disease-associated microglia (DAM)-like programs. Ligand-receptor colocalization analyses identified progranulin (PGRN) as a candidate mediator of these niches. Host- or tumor-cell Grn loss reduced BM burden, and Grn-deficient macrophages exhibited altered metabolic programs and tumor-cell engulfment in vitro. Across independent human BM datasets, GRN expression was associated with conserved lysosomal and DAM-like myeloid programs, and GRN-high myeloid regions colocalized with immunosuppressive signatures and dysfunctional CD8+ T cell states. These cross-species findings identify PGRN as a candidate mediator of the BM immune niche and support further investigation of its therapeutic relevance.

cancer biology↗

bioRxiv2026

Chemically modified CRISPR enzymes for multi-organ genome editing in vivo

Delivery remains the main obstacle to the development of in vivo genome editing therapies. CRISPR ribonucleoproteins confer high editing activity with transient exposure but lack intrinsic cell entry and targeting. Here we introduce PERCEPT, a delivery platform featuring reversible, covalent modification of CRISPR enzymes. PERCEPT enables modular installation of shielding polymers, amphiphilic delivery peptides and targeting ligands, allowing traceless cytosolic release of the native editor. A formulation incorporating an amphiphilic delivery peptide and the neuron-targeting ligand TET1 edited approximately 56% of striatal volume and 78% of neurons within edited regions after local striatal administration. In the R6/2 Huntington's disease model, PERCEPT mediated targeted editing of the mutant human HTT transgene, reducing mutant huntingtin aggregate burden and shifting local transcriptional programs away from inflammatory and injury-associated states. Tissue-adapted formulations edited 53% of Muller glia following intravitreal delivery, increased skeletal muscle reporter fluorescence tenfold versus unconjugated control following intramuscular injection, and enabled lung airway epithelial cell that persisted for three months following intranasal delivery. Rapid screening revealed that local anatomical and cellular barriers require distinct surface display for optimal editing. These results establish reversible chemical modification as a general strategy for adapting CRISPR enzyme delivery across multiple target tissues in vivo.

bioengineering↗

bioRxiv2026

Thal-Kak: unifying biomolecular structure predictors reveals a sampling-selection gap

Complementary all-atom structure predictors sample different solutions, but how to allocate a fixed sampling budget across them and select the best output remains unclear. Thal-Kak unifies five released predictors under shared upstream inputs and a common schema. Across FoldBench and CASP16, model mixing improves oracle sampling over single-model runs, but selection remains a bottleneck because confidence scores do not transfer across models and existing quality-assessment methods cannot resolve this gap.

bioinformatics↗

bioRxiv2026

PINT: Pathway-pathway interactions for predicting interpretable clinical outcomes from gene expression

MotivationDisease mechanisms emerge from the coordinated activity of multiple biological pathways, rather than from individual pathways acting in isolation. Existing pathway-based deep learning models, however, treat pathways as independent entities, aggregating their representations through fully connected layers that disregard inter-pathway relationships. This architectural limitation overlooks an important dimension of disease biology, potentially constraining both predictive performance and the capacity to generate biologically meaningful interpretations. ResultsWe introduce a pathway-based attentive interpretability model, named PINT, that models interactions among pathways through a self-attention mechanism from gene expression data. An attention-based pooling layer further identifies patient-specific pathway contributions to the final prediction. Evaluation across five TCGA cancer datasets demonstrated that PINT consistently outperformed benchmark models in survival analysis. More importantly, PINT identifies pathways significantly associated with survival as well as reveals biologically meaningful interactions among pathways. In the BRCA dataset, PINT identified significant pathways, pathway-pathway interactions, and gene-level contributions within pathways for individual patients, most of which were supported by existing literature. Specifically, the RAS signaling pathway emerged as significantly associated with patient survival, and the learned interaction scores recovered known relationships between RAS signaling and several regulatory pathways, including cAMP, TNF, and Rap1 signaling. Availability and implementationThe source code and data are available at https://github.com/datax-lab/PINT.

bioinformatics↗

bioRxiv2026

Multiplex Genome Editing Overcomes Photoperiod Sensitivity in Tropical Maize

Tropical maize is a rich source of genetic diversity that could enhance temperate maize breeding programs, but its sensitivity to long-day photoperiods, resulting in delayed flowering, limits its widespread use. To overcome this barrier, the Genome Engineering to Sustain Crop Improvement (GETSCI) project used CRISPR/Cas9 to mutate three flowering repressor genes, ZmCCT9, ZmCCT10, and ZmRAP2.7, in the tropical inbred Tzi8. A single sgRNA targeting the first exon of each target gene was combined with an excision cassette carrying the morphogenic genes Babyboom (Bbm) and Wuschel2 (Wus2) to enable efficient transgenic plant regeneration. Transgenic plants carrying frameshift edits in each target gene were recovered, and subsequent crosses produced two non-transgenic genotypes: a double-edited zmcct10, zmrap2.7 line and a triple-edited zmcct9, zmcct10, zmrap2.7 line. Multiple flowering traits were measured for the edited genotypes and unedited Tzi8 inbred in short-day (Hawai i) and long-day (Iowa) field conditions. Both edited genotypes flowered significantly earlier than Tzi8 in both environments. Notably, under long-day conditions, flowering of the two edited lines overlapped with that of the temperate inbred B73, whereas Tzi8 did not. Together, these results demonstrate that targeted, multiplex gene editing can reduce photoperiod sensitivity in a tropical inbred, expanding access to previously untapped genetic diversity for temperate maize improvement.

plant biology↗

bioRxiv2026

Multi-modal Graph Integration for Biologically Interpretable Domain Identification Using Spatial Transcriptomics

Functional domain identification in spatial transciptomics transforms spatial molecular measurements into mechanistic insights into tissue physiology and pathology. However, the inherent noise and sparsity of gene expression data, along with the locality-biased design of conventional graph-based approaches, fundamentally limit the accurate identification of complex tissue domains. In this study, we propose a novel Biologically Interpretable multi-modal Graph using Spatial Transcriptomics, called BIGraph-ST, that integrates pathway activity scores and histological image features for robust spatial domain identification. BIGraph-ST represents modality-specific similarity through affinity graphs and propagates spatial topology to capture higher-order connectivity within the tissue microenvironment. Experimental results demonstrated robust performance and notable improvements across multiple gold-standard benchmark datasets, particularly in cancer tissues. Moreover, BIGraph-ST provides biologically interpretable pathway-level representations of domains, which ultimately offers a valuable tool to gain biological in-sights into complex tissue architectures. The source code will be publicly available upon acceptance.

bioinformatics↗

bioRxiv2026

Recognition Mechanism of Serotonin by a G-Quadruplex-Duplex Hybrid Aptamer

Serotonin is a key neurotransmitter, and aptamer-based tools using the 44-nt Apt44 have been successfully developed for its in vitro and in vivo detection. Nevertheless, the structural basis of recognition by this aptamer remains unclear. Here we report high-resolution NMR structures of Apt38, a 6-nt truncated variant in the third loop of Apt44, in free and serotonin-bound states. Both structures reveal a two-layered antiparallel chair-type G-quadruplex core with three edgewise loops and a terminal duplex, forming a G-quadruplex-duplex hybrid structure. Serotonin binds at the G-quadruplex-duplex junction, stabilized by stacking, electrostatic attraction, hydrogen bonding, and hydrophobic contacts. Apt38 is preorganized for binding, whereas the longer third loop of Apt44 introduces conformational dynamics into the G-quadruplex scaffold, which enables a pronounced binding-triggered conformational switch in PBS buffer, explaining its sensing mechanism. Our work reveals the recognition and sensing mechanism of the serotonin aptamer and provides a framework for aptamer design in serotonin biosensing.

biophysics↗

bioRxiv2026

Cryptic Evolution of Heteroresistance as Adaptation to Treatment Interruptions

The evolution of antibiotic resistance is traditionally understood as a selective sweep to fixation, yielding easily detectable, population-wide resistance. Many clinical isolates, however, exhibit a subtle phenotype in which resistance remains hidden within a susceptible majority despite a clonal genetic background: a phenomenon clinically recognized as heteroresistance (HR). Treatment failure driven by HR has been widely reported across bacterial and fungal infections and in cancer therapy. To understand when and how HR evolves, and why it is selected over classical population-wide resistance, we conducted de novo evolution experiments starting from susceptible Escherichia coli and analyzed the genetic changes and fitness effects in the evolved strains. Prolonged gaps in antibiotic exposure are required for HR to evolve, implicating treatment interruptions as a key driver. HR emerges rapidly and reproducibly with minimal antibiotic use, yet its emergence is not readily detected by routine susceptibility testing. Unlike classical resistance, an evolved HR population partitions at the single-cell level into multiple phenotypes with distinct growth-resistance trade-offs. Their relative abundance shifts dynamically with antibiotic exposure, enabling robust population survival while avoiding the constitutive fitness burden associated with classical resistance. Despite this phenotypic flexibility, stable single mutations including a missense substitution and a short in-frame deletion are sufficient to generate HR, indicating a low evolutionary barrier. Additionally, we found that clinical isolates exhibit genetic and fitness signatures resembling those of our lab-evolved strains, suggesting that clinical HR emerges through the selective mechanism uncovered in our experiments. Together, our results establish HR as a readily evolvable adaptive strategy under treatment interruptions that leverages phenotypic flexibility to maintain resistance at minimal fitness cost, providing mechanistic insight into its emergence and prevalence.

evolutionary biology↗

bioRxiv2026

Metabolic signals regulate resuscitation speed of antibiotic persister bacteria during infection

All living organisms adjust their metabolism in response to environmental changes. Under unfavorable conditions, organisms enter a state of dormancy by halting metabolism, enabling survival. Dormant bacteria become highly tolerant to antibiotics-a phenomenon called persistence. Here, we demonstrate that selective metabolic reprogramming controls the resuscitation speed of persister after antibiotic exposure. Using multi-omics and in silico modeling, we found that dormant bacteria reprogram metabolic pathways to modulate persister awakening. Accumulation of L-serine and reduction of arginine drive rapid resuscitation. L-serine promotes cysteine biosynthesis and motility while reducing energy metabolism to facilitate rapid resuscitation. In contrast, arginine slows regrowth from dormancy by enhancing ethanol-aldehyde and energy metabolism. L-serine and arginine can, respectively, promote or inhibit the regrowth of antibiotic persister cells in macrophages and mouse models, and regulate the awakening speed of Salmonella, E. coli, and methicillin-resistant Staphylococcus aureus (MRSA). These findings suggest new strategies to target chronic bacterial infections. TeaserL-serine speeds and arginine slows the awakening of antibiotic persisters, revealing targets for chronic infection.

microbiology↗

bioRxiv2026

Multi-omics analysis of type II diabetic wound healing reveals CD44-mediated immune cell crosstalk dysfunction in mice and humans

Type II diabetes mellitus (T2DM) is one of the most prevalent diseases in the United States and is associated with diabetic foot ulcers (DFU) and their impaired, often chronic, wound healing. The T2DM mouse model with dysfunctional leptin receptor (db/db) has been used in basic and translational studies of wound healing due to its systemic phenotypes (hyperphagia, hypometabolism, obesity, T2DM) and its notable delayed skin wound healing. However, a characterization of the temporal cellular dynamics of the db/db wound healing model has not been performed, nor has the model been systematically compared to human DFUs. We performed the first comprehensive single-cell, multi-omic analysis of dermal cells in diabetic (db/db) compared to non-diabetic (ND) mice across three time points ranging from the inflammatory to the delayed proliferative and resolution phases of healing. Single-cell transcriptomics were uniquely linked to their corresponding cells surface protein expressions of cell-specific receptors, including immune cells (CD45) such as neutrophils (CD11b, Ly6G), monocytes/macrophages (CD11b, F4/80, CD11c, Ly6C) and T lymphocytes (CD3, CD4), and dermal cells such as endothelial cells (CD31) and fibroblasts (CD26, CD140a), and showed high concordance between protein cell markers and their gene expressions in major cell types. Differential multi-omic analyses characterized two neutrophil (Tnfaip3+Sod2+Ly6G+, Csf3r+Fos+Ly6G+), three monocyte/macrophage (F4/80highCD11bhigh, Ly6chighCD11bhigh, CD11chighCD11blow) and three fibroblast (Pi16+Dpp4+CD26high, Lrrc15+Tnc+CD140ahigh, Cilp+Mgp+CD26low) subtypes showing dysregulated dynamics across the time course of healing in db/db vs ND mice. Notably, NETotic Tnfaip3+Sod2+Ly6G+ neutrophils and phagocytic F4/80highCD11bhigh macrophage subtypes were drastically up-regulated in diabetic wounds. Differential cell-cell communication analyses revealed striking differences in crosstalk dynamics between fibroblast, macrophage and neutrophil subtypes in the early phase of healing, and ligand-receptor interactome analyses identified CD44 as the hub of dysregulated immune cell interactions in diabetic wounds, implicating cell adhesion, migration and inflammatory pathways, especially those mediated by ICAM1. Inhibition of CD44 using blocking antibodies in primary macrophages from db/db mice and via intradermal injections in db/db mice significantly normalized the early wound immune dysfunction, in part by inhibiting ICAM1 and reversing the excessive neutrophil influx into diabetic wounds. A new integrated dataset of single-cell human chronic wound studies revealed similar CD44-mediated immune cell dysfunctions in diabetic vs non-diabetic foot ulcers, pointing to CD44 as a promising therapeutic target for T2DM-associated chronic wounds.

molecular biology↗

bioRxiv2026

An auxiliary protein tunes reductase activity in alkaloid biosynthesis

Monoterpene indole alkaloids encompass a diverse class of plant natural products. Many of these valuable compounds, including the anticancer medicines vinblastine and vincristine (Catharanthus roseus) and the antiaddiction treatment ibogaine (Tabernanthe iboga) utilize a common biosynthetic intermediate, 19E-geissoschizine, which is produced by the medium-chain dehydrogenase/reductase geissoschizine synthase (GS). Herein we report the discovery of a non-catalytic auxiliary protein (facilitator of geissoschizine synthase, FoGS) that improves formation of 19E-geissoschizine by acting in combination with GS. The discovery of a FoGS orthologue that works with GS to produce 19Z-geissoschizine shows how this protein also controls the stereochemistry of the product. We demonstrate that FoGS and GS interact to form a stable heterodimer, and show through mutagenesis that FoGS likely modulates the structure of the GS active site. Microscopy studies suggest that FoGS also changes the subcellular localization of GS. The discovery of FoGS will enable efforts to heterologously reconstitute commercially important geissoschizine-derived monoterpene indole alkaloids, and moreover, highlights how metabolic pathways utilize proteins that are not essential for production of the products, but nevertheless enhance the efficiency and specificity of the catalytic biosynthetic pathway components.

biochemistry↗

bioRxiv2026

Conserved metabolic vulnerabilities across pathogenic coronaviruses nominate host-directed therapeutic targets

Pathogenic coronaviruses profoundly rewire host cell metabolism to support viral replication, yet whether these metabolic alterations expose shared and actionable vulnerabilities remains unclear. By integrating transcriptomic profiles from cells infected with SARS-CoV, SARS-CoV-2, and MERS-CoV with genome-scale metabolic models, we identify conserved and virus-specific metabolic perturbations affecting mitochondrial transport, nucleotide biosynthesis, fatty acid metabolism, and redox balance. Despite distinct transcriptional responses, all three viruses converge on a limited set of metabolic reactions whose flux ranges deviate strongly from healthy states. Using a network-based predictive framework, we systematically identify gene-pair perturbations that restore perturbed reaction fluxes toward non-infected metabolic states. Predicted rescue mechanisms reveal shared metabolic dependencies across coronaviruses, as well as time-dependent virus-specific vulnerabilities, and nominate druggable host targets. Notably, several top predictions align with independent experimental and clinical evidence, including metabolic interventions shown to reduce viral replication or disease severity in COVID-19 patients. Together, our results define conserved metabolic rescue pathways in coronavirus infection and provide a general strategy for identifying host-directed therapeutic opportunities from transcriptomic data. HighlightsO_LICoronaviruses converge on shared metabolic vulnerabilities in host cells C_LIO_LINiTRO predicts gene pairs that rescue viral-induced metabolic states C_LIO_LIMitochondrial transport emerges as a key pan-coronavirus target C_LIO_LITop predictions validated by clinical trials and in vitro evidence C_LI eTOC BlurbDohai et al. develop NiTRO, a network-based algorithm that integrates coronavirus-induced transcriptomic changes with genome-scale metabolic models to identify gene-pair perturbations capable of rescuing infected metabolic states. The approach reveals shared and virus-specific druggable metabolic vulnerabilities, with top predictions corroborated by clinical evidence.

systems biology↗

bioRxiv2026

Benchmarking three simple DNA staining-based image metrics for live-cell tracking of chromatin organization

Quantifying chromatin-state dynamics in living cells remains challenging, in part because most methods require fixation or cell lysis. Here, we benchmark and introduce three simple live-cell image-derived metrics computed from routine DNA staining - the coefficient of variation (CV), 1-Gini, and the Diffuse Signal Index (DSI), introduced here - as fixation-free readouts of chromatin organization. Using HL60-derived neutrophils (dHL-60) undergoing NETosis as a model system with a pronounced compact-to-decompact chromatin transition, we show that all three metrics track progressive chromatin reorganization in live-cell trajectories, with DSI providing the strongest trajectory-level discrimination between NETing and non-NETing cells. Comparison with Tn5-based chromatin accessibility measurements in fixed cells shows that all three metrics correlate with chromatin accessibility, supporting their biological interpretability. We further show that all three metrics track chromatin reorganization in a second, mechanistically distinct biological process, capturing mitotic chromatin compaction and post-mitotic chromatin decompaction in U2OS cells. Together, our results provide a practical framework for extracting chromatin-organization readouts from routine live-cell DNA staining and highlight that metric performance is task- and context-dependent. To make our metrics accessible to a broad audience, we built NucMetrics, an open-source ImageJ/Fiji macro toolset that computes all three metrics without requiring programming.

cell biology↗

bioRxiv2026

Ca2+ influx through ER-plasma membrane contacts is required for brown fat thermogenesis and metabolic health

Brown adipose tissue (BAT) exhibits exceptional metabolic plasticity, rapidly increasing energy expenditure to sustain thermogenesis during cold exposure. This high metabolic activity imposes substantial demands on cellular systems, requiring robust adaptive mechanisms to maintain homeostasis and prevent cellular stress. Yet, the pathways that support and coordinate these adaptive responses in brown adipocytes remain incompletely understood. Here, we identify a cold-induced adaptive program in BAT characterized by the formation of endoplasmic reticulum-plasma membrane (ER-PM) contact sites and the activation of store-operated calcium entry (SOCE), which is essential for maintaining brown adipocyte health during thermogenic activation. Cold exposure enhances ER-PM contacts and upregulates the expression of STIM and Orai proteins, key mediators of SOCE. Loss of STIM in brown adipocytes disrupts intracellular Ca{superscript 2} homeostasis and induces aberrant aggregation of ER membranes. STIM deficiency also impairs cold-induced mitochondrial fission resulting in hyperfused mitochondria with reduced oxidative capacity, independently of UCP1 abundance. Importantly, mice lacking STIM in BAT exhibit impaired lipid oxidation, are cold intolerant and develop exacerbated peripheral insulin resistance when challenged with a high-fat diet. Together, these findings identify ER-PM remodeling and STIM-mediated SOCE as a central regulator that links organelle architecture to brown adipocyte function and contributes to whole-body metabolic homeostasis.

cell biology↗

bioRxiv2026

CEACAM5/6+ Tumor Cells and IL-1β+ Macrophages Drive Resistance to Chemo-immunotherapy in Gastric Cancer

Chemo-immunotherapy is a first-line treatment for advanced gastric cancer, yet response rates remain limited and resistance mechanisms are poorly defined. Here we generate a single-cell atlas of 542,121 cells from 35 patients treated with anti-PD-1 plus chemotherapy, profiling pre- and post-treatment tumors linked to clinical response. Integrating spatial transcriptomics, immunohistochemistry, and bulk RNA sequencing, we identify two temporally distinct resistance programs. Intrinsic resistance in pre-treatment non-responders is marked by enrichment of CEACAM5/6 tumor cells that form immune-excluded spatial niches characterized by macrophage recruitment and CD8 T-cell exhaustion. Acquired resistance in post-treatment non-responders is driven by expansion of IL-1{beta} macrophages, which induces coordinated NF-{kappa}B activation across tumor and stromal compartments, promoting PD-L1 upregulation, epithelial-mesenchymal transition, and chronic inflammation. These findings delineate an evolutionary trajectory of resistance and nominate CEACAM5/6 and IL-1{beta} as predictive biomarkers and therapeutic targets to improve anti-PD-1-based combination strategies.

cancer biology↗

bioRxiv2026

Functional Locality-Aligned Learning Reveals Structure-Function Causality in Enzyme Kinetics

Accurate estimation of enzyme kinetic parameters is essential for enzyme engineering and industrial biocatalysis, yet their experimental measurement remains labor-intensive and costly. Although machine learning offers an efficient alternative, existing methods still struggle to generalize to unseen enzymes and substrates. In particular, current three-dimensional (3D) structure-aware approaches rely on whole-enzyme 3D geometric structures and neglect substrate 3D geometry, often yielding limited or even degraded performance. We identify a fundamental limitation underlying these methods: a mismatch between the structural representation learning scale and the functional locality scale, which weakens structure-function causality in enzyme kinetics. To address this issue, we introduce EnzymePlex, a functional locality-aligned framework that aligns inductive biases with the localized structural determinants of enzyme function by prioritizing catalytic pockets, integrating substrate 3D geometry, and modeling nuanced enzyme-substrate interplay under the guidance of pocket-level structural priors. EnzymePlex achieves state-of-the-art performance across multiple benchmarks and substantially improves generalization under stringent out-of-distribution evaluations. Beyond predictive accuracy, EnzymePlex learns mechanistically aligned representations, with attention enriched at catalytic pocket residues and substrate reaction centers despite receiving no explicit super-vision for either. Moreover, when applied to recently reported wet-lab data, EnzymePlex effectively prioritizes high-activity enzyme variants and identifies potent inhibitors, highlighting its potential to accelerate enzyme engineering and drug discovery.

biochemistry↗