Biology subjects

Tan, H.

Publications and source records attributed to Tan, H..

At least 19 recordsLinked to original sources

bioRxiv2026

A lifespan-scale single-cell atlas defines an early-childhood immunometabolic transition linked to age-referenced immune states

Childhood represents a critical period with rapid metabolic and immune remodeling, during which perturbations may have lasting consequences for health. We integrated single-cell RNA sequencing (scRNA-seq) profiles of peripheral blood mononuclear cells (PBMCs) from 2,008 healthy individuals aged 0-100 years. Age-resolved modeling highlighted an early-childhood transition centered at age 6, characterized by coordinated remodeling of metabolism-related transcriptional programs, particularly in monocyte and B-lineage cells. Matched serum metabolomics and lipidomics in a pediatric subcohort detected a temporally concordant wave of metabolite and lipid changes around ages 6-8 and extensive metabolite/lipid-transcript associations, prominently involving steroid and phospholipid species. Building on these findings, we developed an immunometabolic transition score and age-specific reference chart, which were evaluated in an independent healthy cohort and applied to quantify deviations in pediatric immune disorders. These findings highlight a coordinated immunometabolic transition in early childhood and provide an age-referenced framework for distinguishing atypical immune-state variation from healthy age-associated variation.

immunology↗

bioRxiv2026

Factorial Knockouts Distinguish Physical Necessity from Numerical Compensation in a PSI-LHCI Transport Model

Mechanistic interpretation of photosynthetic energy-transfer models requires more than reproduction of experimental observables: a model intended to support mechanistic claims should also respond consistently when its proposed functional organizations are removed. Here, we evaluate whether a calibrated PSI-LHCI transport surrogate encodes physically meaningful principles by applying a 2x2 factorial knockout framework that independently removes site-energy heterogeneity and coupling-strength heterogeneity in a 155-pigment network. All perturbations were evaluated using the same Full-model calibration without parameter refitting. Although the calibrated model reproduced a high excitation-trapping yield, eliminating either energetic or coupling heterogeneity unexpectedly improved its apparent transport performance. A strict zero-coupling control confirmed that coupling itself remained necessary for network-mediated reaction-center access, whereas the supplied organization of coupling strengths was not supported by the surrogate. An audit of the model inputs further identified peripheral localization of all lowest-energy states and effective coupling scales far above those used in structure-based chlorophyll Hamiltonians. These findings do not imply that native PSI favors flat energy landscapes or uniform couplings. Instead, they show that endpoint agreement alone does not validate a mechanistic interpretation of a pigment-network model. Factorial knockout analysis provides a falsification-oriented framework for separating physical necessity from proposed organization, diagnosing numerical compensation, and identifying the constraints required for more predictive models of PSI-LHCI energy transfer.

biophysics↗

bioRxiv2026

Trained phages constrain evolutionary routes to antibiotic resistance in Escherichia coli

Bacteriophages are major drivers of bacterial evolution and can impose strong selection on traits that mediate resistance and fitness. This creates the possibility of using phage to steer microbial adaptation toward predictable phenotypic outcomes. Yet bacterial populations often evolve phage resistance through diverse genetic routes, not all of which impose meaningful costs or trade-offs. Such evolutionary flexibility can allow bacteria to escape phage selection without acquiring the desired phenotype, limiting the reliability of phage-based evolutionary steering. Here, we develop a directed phage evolution strategy that eliminates cost-free resistance pathways, forcing bacteria into evolutionary trajectories that restore antibiotic susceptibility. We isolated two novel Microviridae phages (BLS2 and BLS5) that infect E. coli BL21 and found that bacteria evolved resistance through two distinct mechanisms: mutations in lipopolysaccharide (LPS) biosynthesis genes, which carried substantial fitness costs and conferred sensitivity to erythromycin, and loss-of-function mutations in yajC, which imposed no fitness penalty and maintained full antibiotic resistance. We show that YajC is a previously uncharacterised accessory receptor required for phage genome translocation across the inner membrane. We then used this mechanistic understanding to design a phage training regime that evolved phages capable of infecting yajC mutants, thereby closing this cost-free escape route. Bacteria challenged with trained phages were constrained to evolve LPS-based resistance, and were consequently resensitised to erythromycin. Our results demonstrate that directed phage evolution can reshape the fitness landscape of bacterial resistance, channelling adaptation along trajectories that carry predictable phenotypic costs - with broad implications for phage-mediated manipulation of microbial populations. Graphical Abstract O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=156 SRC="FIGDIR/small/742405v1_ufig1.gif" ALT="Figure 1"> View larger version (39K): org.highwire.dtl.DTLVardef@14b387aorg.highwire.dtl.DTLVardef@168ff62org.highwire.dtl.DTLVardef@56a8c9org.highwire.dtl.DTLVardef@16363e5_HPS_FORMAT_FIGEXP M_FIG C_FIG

evolutionary biology↗

bioRxiv2026

AMPKγ2 Deacetylation Drives Nuclear Translocation and Doxorubicin-Induced Cardiomyopathy via Nucleolar Stress Signaling

Doxorubicin (Dox)-induced cardiomyopathy (DIC), characterized by cardiomyocyte apoptosis, remains a major clinical challenge in chemotherapy. The regulatory {gamma}2 subunit of AMP-activated protein kinase (AMPK{gamma}2) plays a key role in cardiovascular diseases, but its function in DIC is poorly understood. Here, we report that Dox induces isoform-specific deacetylation and nuclear accumulation of {gamma}2, triggering nucleolar stress and p53-mediated apoptosis. Mechanistically, HDAC3 and TIP60 interact with {gamma}2 and modulate the acetylation of multiple lysine residues within its nuclear localization signal (NLS), controlling its nucleocytoplasmic shuttling. Dox enhances HDAC3-mediated {gamma}2 deacetylation, thereby driving nuclear accumulation of the {gamma}2-containing AMPK ({gamma}2-AMPK) while suppressing the cytosolic AMPK activity. Nuclear {gamma}2-AMPK phosphorylates and inactivates TIF-IA, a key RNA polymerase I-specific transcription initiation factor, leading to nucleolar stress through inhibition of rRNA transcription. rRNA deficit triggers release of free ribosomal proteins (RPs), which bind to and inhibit the E3 ubiquitin ligase MDM2, resulting in p53 stabilization and activation of apoptotic signaling. Using genetically engineered cardiomyocytes and a DIC mouse model, we found that a deacetyl-mimetic {gamma}2 mutant (6KR) exacerbated DIC, whereas an acetyl-mimetic mutant (6KQ) was cardioprotective. Collectively, our findings establish acetylation-driven nuclear translocation of {gamma}2 as a critical node linking Dox-induced nucleolar stress to p53-dependent apoptosis and suggest a promising cardio-oncology strategy that combines HDAC inhibitors with Dox to mitigate DIC. Significance statementDoxorubicin is an effective cancer drug, but its use is limited by cardiomyopathy. Our study reveals that doxorubicin drives HDAC3-mediated deacetylation of AMPK{gamma}2, exposing its nuclear localization signal and redirecting {gamma}2-containing AMPK from the cytoplasm to the nucleus. Nuclear AMPK{gamma}2 phosphorylates TIF-IA, suppresses ribosomal RNA synthesis, and activates a nucleolar stress pathway that stabilizes p53 and promotes cardiomyocyte apoptosis. In mice, a deacetylation-mimetic AMPK{gamma}2 mutant worsens doxorubicin-induced cardiomyopathy, whereas an acetylation-mimetic mutant is protective. These findings uncover an acetylation-controlled spatial switch in AMPK signaling and identify the AMPK{gamma}2 deacetylation-nucleolar stress axis as a potential target for reducing chemotherapy-associated cardiac injury.

cell biology↗

bioRxiv2026

Device-embedded accelerometry complements neural signals for tracking parkinsonian motor states

Adaptive deep brain stimulation (aDBS) relies on physiological biomarkers to infer motor state and guide therapeutic stimulation in Parkinsons disease. However, neural biomarkers may themselves be altered by stimulation, potentially limiting their utility for closed-loop control. We address this limitation by testing whether DBS device-embedded accelerometers can accurately track Parkinsonian motor state across stimulation conditions. We analysed over 1,900 hours of chronic recordings of subthalamic nucleus (STN), sensorimotor cortical and device-embedded accelerometry signals acquired before and during continuous STN stimulation, alongside continuous wearable assessments of bradykinesia and dyskinesia. Across stimulation conditions, accelerometry-derived features robustly tracked motor symptom severity and outperformed neural features for symptom decoding. Mechanistically, total STN beta power - a widely used biomarker for aDBS - proved less informative because it conflates periodic and aperiodic neural processes with opposing relationships to motor state. Under active stimulation, periodic beta activity showed reduced coupling to symptom severity, whereas STN aperiodic activity, cortical periodic activity and cortico-subthalamic coherence remained comparatively stable. Together, these findings demonstrate that neural and behavioural biomarkers exhibit differential robustness during deep brain stimulation and identify device-embedded accelerometry as a robust behavioural biomarker of motor state, motivating its use in next-generation adaptive DBS systems.

bioengineering↗

bioRxiv2026

Reinforcement learning-driven unified generative framework for multi-objective RNA codon design

Current RNA codon design methods are limited by inefficient long-sequence processing and poor generalizability, often relying on a decoupled generate-or-optimize paradigm. We introduce RNARL, a reinforcement learning-driven framework that unifies sequence generation with multi-objective optimization. RNARL directly learns to generate high-performance sequences, effectively optimizing sequences over 3,900 nucleotides and demonstrating superior performance and universality across six species and five RNA types. RNARL thus establishes an effective and generalizable framework for RNA codon design. Finally, a user-friendly web platform is freely available to facilitate its application for RNA therapeutic design.

bioinformatics↗

bioRxiv2026

GalaxyVS: Exploring 100-Billion Compounds in Seconds

We present GalaxyVS, a hardware-software co-designed virtual screening framework built to explore the 100-billion commercially accessible chemical space in seconds, deployed at the National Supercomputing Center in Tianjin. Built upon the dense vector retrieval paradigm of DrugCLIP, GalaxyVS bypasses the structural dependencies and computational overhead of classical docking to enable rapid screening against experimentally determined as well as geometrically feasible pockets on AlphaFold-predicted structures. To scale this paradigm to the 100-billion level, the system must overcome the significant computational burden of offline representation encoding, critical memory and I/O bottlenecks during online retrieval, and the risks of diversity collapse and precision loss within final screening results. Utilizing the heterogeneous supercomputing infrastructure, GalaxyVS accelerates the offline encoding through deep operator adaptations and resolves online retrieval bottlenecks via disk-native vector indexing coupled with in-memory staging to ensure both broad accessibility and high throughput. Concurrently, a two-stage refinement protocol effectively mitigates diversity collapse and ensures high-fidelity affinity ranking. Consequently, GalaxyVS achieves a daily scoring throughput of 1.5 x 1016 target-ligand pairs, representing a six-orders-of-magnitude leap over previous supercomputing records. Driven by this throughput, we screened nearly 100,000 protein structures across six species against the 100-billion compound library in just 16 hours. The resulting comprehensive cross-species interaction landscape, GalaxyDB, will be openly released at https://galaxyvs.drugclip.com.

bioinformatics↗

bioRxiv2026

Genetic suppression of myeloid receptor Clec7a attenuates microglia neuroinflammation and promotes microglial phagocytosis to delay disease progression in ALS models

Microglial activation has been closely associated with accelerated ALS disease progression. However, specific microglial pathways that regulate microglial activation and ALS disease progression remain limitedly understood. Here, we determined the role of Clec7a (or Dectin-1), a core signature gene of disease-associated microglia (DAM) in ALS, in regulating microglial activation and ALS disease progression. Our spinal cord scRNA-Seq results found that Clec7a deficiency specifically attenuated microglial neuroimmune gene expression in SOD1G93A mice and human ALS. In addition, in vivo two-photon imaging of human (h) TDP43 phagocytosis by microglia in the cortex showed that Clec7a deficiency promotes microglial phagocytosis of pathological hTDP43 by enhancing microglial process dynamics. Subsequent survival analysis further showed that selective deletion of Clec7a in microglia mitigates motor neuron degeneration and delays disease progression in SOD1G93A ALS mice. Together, our results establish that Clec7a is a key regulator in shaping disease microglial functions and promotes disease progression in ALS.

neuroscience↗

bioRxiv2026

APOE is a presynaptic protein that accumulates with age and modulates neurotransmitter release

The synaptic vesicle (SV) cycle is the fastest membrane trafficking and protein sorting process in biology. It underlies neuronal communication and cognition, yet synaptic function declines during normal aging, increasing vulnerability to neurologic disease. How the SV cycle is maintained across the lifespan of a complex organism remains unclear. Here, we used wild-type mice (C57BL/6J) to define the age- and sex-stratified molecular landscape of SVs and identified apolipoprotein E (APOE) as an abundant presynaptic protein further enriched in aged female samples. Super-resolution imaging, cell-type selective expression, and protease protection assays demonstrate that APOE originates from astroglia and associates with the cytosolic face of SVs. Using iGluSnFR and pHluorin optophysiology, we find that both decreased and increased APOE levels impair neurotransmission during stimulus trains. Together, these findings place APOE at the synapse and establish it as a cell-nonautonomous regulator of the SV cycle.

neuroscience↗

bioRxiv2026

Connectome lateralization in autism across the first 14 years: heterogeneity related to developmental stage, hemisphere, and pathophysiology

Hemispheric lateralization is fundamental feature to neurodevelopment, yet its dynamic evolution in autism across childhood and underlying molecular mechanisms remain unknown. Thus, using a large-scale resting-state fMRI dataset (n = 1553, ages 1-14), we investigated developmental heterogeneity in connectome lateralization. At the group level, abnormalities in dynamic connectome lateralization strength (DCLS) progressed from focal patterns in early childhood ([≤] 6 years) to widespread distributions by late childhood ([≤] 14 years), independent of symptom severity. The late-stage patterns were validated using the ABIDE dataset (n=153). Late childhood also showed marked increases in inter-individual heterogeneity. These alterations were significantly associated with clinical phenotypes, transcriptomic profiles, and multiple neurotransmitter systems, with distinct patterns across childhood groups. Our findings reveal a developmentally dynamic trajectory of connectome lateralization in autism and highlight late childhood as a critical window for stage- and individual-specific interventions tailored to substantial individual heterogeneity.

neuroscience↗

bioRxiv2026

Single Plaque Proteomics Reveals the Composition and Dynamics of the Amyloid Microenvironment in Alzheimer's Disease

Alzheimers disease (AD) is characterized by amyloid plaques that form complex microenvironments in the brain. However, the molecular composition of these plaques and their temporal regulation are not well defined. Here, we developed a sensitive workflow for quantitative proteomic profiling of single plaques using refined laser capture microdissection and data-independent acquisition mass spectrometry (LCM-DIA-MS). From >200 plaques and control regions in AD mouse models (5xFAD and APP-KI) and human brains, we quantified >7,000 proteins, revealing stage-dependent, cell-type-related remodeling of the amyloid proteome (amyloidome). Temporal profiling uncovered early immune and lysosomal activation followed by engagement of RNA processing and synaptic pathways. Cross-model and cross-species analyses determined a conserved amyloidome including APOE, MDK, PTN, and HTRA1, validated by co-localization in imaging analysis. Network analysis highlighted modules in lipid transport, vesicle organization, and autophagy. These findings establish amyloid plaques as conserved, dynamic multicellular hubs that link amyloid accumulation to downstream cellular events.

neuroscience↗

bioRxiv2026

Driving theta-gamma oscillations modulates extrasynaptic GABAergic tone: a tACS-TMS study

BACKGROUNDTheta-gamma phase-amplitude coupled ({theta}{gamma}-PAC) oscillations in primary motor cortex (M1) have been shown to support motor skill acquisition. Past research has shown that driving gamma activity at the theta peak (TGP), but not the theta trough (TGT) using transcranial alternating current stimulation (tACS) enhances motor learning (Akkad et al., 2021). However, the neurophysiological mechanisms underlying this phase-specific effect remain unclear. METHODSIn a double-blind, sham-controlled, cross-over study, twenty-two healthy participants received 20 minutes of 75Hz/6Hz TGP-tACS, TGT-tACS, or sham stimulation over M1. We used paired-pulse transcranial magnetic stimulation (TMS) to assess GABAergic and NMDAR-mediated activity before, during, and after tACS. Outcome measures included short-interval intracortical inhibition at 1ms (SICI1ms; extrasynaptic GABAergic tone) and 2.5ms (SICI2.5ms; synaptic GABAA activity), intracortical facilitation at 12ms (ICF12ms; NMDAR activity), and motor evoked potential (MEP) amplitude (corticospinal excitability). RESULTSTGP-tACS selectively decreased SICI1ms, a putative marker of extrasynaptic GABAergic tone (main effect of Stimulation: p=.021), with significant differences between TGP and TGT during late stimulation (p=.047). No significant effects were observed on corticospinal excitability, synaptic GABAergic activity (SICI2.5ms), or NMDAR signalling (ICF12ms). CONCLUSIONSDriving theta-gamma oscillations at the theta peak using tACS specifically modulates extrasynaptic GABAergic tone in M1 without affecting corticospinal excitability or synaptic inhibition. Given that reductions in GABAergic signalling supports motor learning, these findings provide a neurophysiological mechanism for the phase-specific behavioural effects of {theta}{gamma}-PAC tACS and suggest a potential therapeutic approach for facilitating motor recovery after stroke. HighlightsO_LITheta-gamma peak tACS selectively reduces extrasynaptic GABA in human motor cortex C_LIO_LIOnly gamma at the peak, not trough, of theta stimulation modulates GABA C_LIO_LINo effects on corticospinal excitability, synaptic GABA, or NMDAR signalling C_LIO_LItACS-TMS reveals mechanism for phase-dependent motor learning effects C_LI

neuroscience↗

bioRxiv2026

Distinct roles of the human cuneiform and pedunculopontine nuclei in gait initiation and freezing of gait

Freezing of gait in Parkinsons disease (PD) is a major cause of disability, often resistant to dopaminergic therapy and deep brain stimulation (DBS). Its underlying mechanisms remain unclear, mainly because the roles of the human mesencephalic locomotor region (MLR) nuclei are not well understood. Here, we combined rare local field potentials (LFP) recordings from the cuneiform (CuN) and pedunculopontine nuclei (PPN) with biomechanical markers of gait initiation (GI) in four PD patients. We identified functional differences: increases in CuN alpha-band activity precede anticipatory postural adjustments (APA) and correlate with the rhythm of upcoming steps, whereas decreases in PPN beta-band activity occur during APA just before the lead foot lifts off. Imminent freezing is characterized by a breakdown of this organization, marked by mistimed alpha-band surges across the MLR and abnormal PPN beta-band modulation. CuN stimulation selectively improved the stepping rhythm, while PPN stimulation worsened pace or forward vigor. Furthermore, exaggerated mesencephalic alpha-band power was associated with poor clinical responses. These results clarify the individual roles of MLR nuclei in human locomotion and identify pathological alpha dynamics as a biomarker for advancing adaptive neurostimulation.

neuroscience↗

bioRxiv2025

Disentangling motor and cognitive roles of cortical beta oscillations and their modulation with rTMS

Cortical beta oscillations are central to motor control. These rhythms are modulated during movement preparation and execution, and are also sensitive to cognitive factors such as uncertainty, attention and reward prediction. However, the distinct motor and cognitive roles of cortical beta oscillations --roles that may inform condition-specific therapeutic strategies--remain unclear. This study aimed to disentangle the respective contributions of motor and cognitive functions, and to assess their modulation with repetitive transcranial magnetic stimulation (rTMS). Twenty-four healthy participants performed a visually cued reaching task in which directional uncertainty was dynamically manipulated by varying the number of targets in real-time. Reaction time was quantified as a behavioural measure. Three stimulation conditions were applied during the preparatory periods between the uncertainty and go cues: no stimulation, regular rTMS at each participants individual beta frequency, and irregular rTMS. Electroencephalography was used to measure beta-band oscillatory activity. Our results showed that uncertainty cues induced bilateral beta suppression, with greater uncertainty linked to smaller reductions in beta power. Movement-related beta modulation was lateralised to the hemisphere contralateral to the executing hand, where elevated beta power in the pre- go cue period predicted longer reaction times. Both regular and irregular rTMS significantly shortened reaction times and attenuated beta event-related desynchronisation. Reductions in beta desynchronisation correlated with improvements in reaction times, suggesting beta desynchronisation reflects a neural threshold for movement initiation. These findings indicate that cortical beta oscillations encode distinct motor and cognitive processes, and that beta frequency rTMS can modulate beta dynamics to facilitate faster movement initiation by lowering this neural threshold.

neuroscience↗

bioRxiv2025

Dithering suppresses half-harmonic neural synchronisation to photic stimulation in humans

While entraining neural rhythms using brain stimulation has been suggested as a therapeutic mechanism to normalise brain activity in conditions such as depression, chronic pain, or Alzheimers disease, periodic stimulation can also inadvertently entrain brain rhythms at sub- and superharmonics of the stimulation frequency, which could lead to deleterious effects. Slightly jittering stimulation pulses (called "dithering") was previously proposed on the basis of mathematical modelling to selectively entrain a target neural rhythm while avoiding harmonic entrainment. In this study, we investigated the potential of dithering in humans. Using photic stimulation (light flicker) and EEG recordings in healthy participants, we showed that dithering suppresses half-harmonic synchronisation relative to perfectly periodic flicker, and more so than synchronisation at the stimulation frequency. This was also the case for a periodic condition with reduced stimulation amplitude, as predicted by theory. Furthermore, we demonstrated using synthetic data and modelling that the half-harmonic responses observed in participants cannot be explained by the super-position of evoked responses (even when modulated at the half-harmonic frequency), and are better matched by a minimal oscillator model. Our findings are consistent with half-harmonic EEG synchronisation in response to photic stimulation predom-inantly reflecting half-harmonic entrainment rather than the summation of evoked responses, and with dithering being an effective strategy to suppress subharmonic entrainment.

neuroscience↗

bioRxiv2025

Visible Light-Controllable Surgical Hemostatic Adhesives Enabled by Vitamin B12-Stabilized Poly(α-Lipoic Acid)

Water-resistant adhesives are indispensable for biomedical applications, from surgical wound closure to internal hemostasis. While poly(-lipoic acid)-based adhesives show promise for their strong underwater adhesion, their clinical utility has been limited by uncontrolled polymerization-depolymerization dynamics. Here we present a visible light-controllable adhesive system enabled by a simple mix of -lipoic acid (LA) and vitamin B12 (i.e., adenosylcobalamin or AdoB12)--a photolabile compound--without resorting to any chemical modification. Benign visible light irradiation facilitates the cleavage of the C-Co bond and the subsequent formation of B12-thiolate complexes that stabilize LA polymers against depolymerization. The photoresponsive LA-AdoB12 adhesive system has proven highly effective in a wide range of surgical applications, achieving (1) strong bonding in porcine skin models in vitro, (2) reliable sealing of punctured organs (lung, heart and stomach) ex vivo, and (3) immediate hemostasis in active bleeding models such as topical, esophageal, and intestinal wounds in vivo. This work highlights a simple yet powerful strategy for creating visible light-controllable bioglues well-suited for diverse surgical applications.

bioengineering↗

bioRxiv2025

Real-time in vivo imaging of rodent brain response to hypoperfusion enabled by remotely controlled elastomeric micro-balloons

Current preclinical models of ischemic stroke in mice do not permit simultaneous and continuous in vivo brain imaging during the peri-stroke period, therefore missing critical pathophysiological events that could be pivotal for stroke management at the early stage. Here we report remote control of the blood flow of cerebral arteries in live mice continuously at different states to induce stroke in a precise, reliable, and reversible manner. The sub-millimeter scale micro-balloons can expand more than four times their initial diameter, featuring a monolithic elastomeric wall with selectively stiffened regions for controlled inflation and elasticity depending on the target vessels. By allowing for the control of common carotid artery diameter through a cuff pressing against the artery, the micro-balloon recapitulates clinically relevant haemodynamics in a mouse model of global brain ischemia, evidenced by real-time imaging of the brain through intravital microscopy and magnetic resonance imaging. The presented micro-balloons hold significant potential to improve the treatment of stroke patients for minimally invasive intervention and in vivo imaging of the pathophysiological events in the peri-stroke phase.

bioengineering↗

bioRxiv2025

ProStab: Prediction of protein stability change upon mutations by protein language and inverse folding models

Predicting protein stability change upon mutation is critical for protein engineering, yet remains limited by the modeling assumptions of physics-based methods and the generalization bottlenecks of data-driven approaches. We present ProStab, a deep learning framework that integrates sequence- and structure-based information, including the mutation-aware sequence embeddings from protein language models and the geometric features extracted via an inverse folding model. Trained on the large-scale Megascale dataset, ProStab demonstrates strong performance across diverse test sets and robust generalization across distribution shifts between the training and test sets. In head-to-head comparisons, ProStab outperforms all state-of-the-art methods with consistently higher Spearman correlation and precision. To evaluate its practical utility, we experimentally validated ProStab-predicted mutations on the model enzyme transaminase. Among the 16 successfully expressed variants, 4 exhibited improved thermal stability. Remarkably, the 1st top-ranked predicted mutation yielded the highest observed enzymatic activity, retaining three-fold that of the wild type after 10 minutes at 40 {degrees}C. To facilitate broader application, a publicly accessible web server has been developed. We envisage that ProStab provides a scalable and accurate platform for intelligent protein stability design.

bioinformatics↗