Biology subjects

Saito, Y.

Publications and source records attributed to Saito, Y..

At least 19 recordsLinked to original sources

bioRxiv2026

Developmental prefrontal myelination shapes inhibitory interneuron function and adult social behavior

Early-life adversity and abnormal myelination are implicated in psychiatric disorders, but how developmental myelination influences adult social cognition remains unclear. Here, we tested whether disrupting developmental myelination in the medial prefrontal cortex (mPFC) impairs adult social behavior. Using complementary genetic strategies in male mice, we suppressed Myelin Regulatory Factor (Myrf)-dependent myelination locally in the juvenile mPFC or broadly in oligodendrocyte-lineage cells. Both approaches selectively impaired adult social novelty preference while sparing sociability, locomotion, and affective behaviors. Focal prefrontal dysmyelination also reduced social dominance. Spatial transcriptomics revealed coupling between myelin-related genes and parvalbumin (PV) interneuron markers in the healthy mPFC, which was disrupted by developmental hypomyelination. Consistent with this transcriptional alteration, PV interneurons showed reduced high-frequency firing despite preserved intrinsic excitability. In the human prefrontal cortex, MBP-PV expression was correlated in controls but not in individuals with schizophrenia. These findings indicate that developmental prefrontal myelination contributes to PV interneuron maturation and suggest that disrupted oligodendrocyte-interneuron coordination may contribute to impaired social recognition.

neuroscience↗

bioRxiv2026

Engineering a highly active thermophilic F1-ATPase by homolog-guided exploration and machine-learning-assisted prioritization

The rotary motor F1-ATPase has been extensively studied as a model molecular machine, yet rational engineering of its catalytic activity remains challenging because ATP hydrolysis is regulated by long-range intersubunit allostery and large conformational transitions. Here, we developed a homolog-guided engineering strategy to increase the maximum rotation rate of the thermophilic Bacillus PS3 F1-ATPase (TF1). Candidate mutation sites were first identified by comparing TF1 with the homologous enzymes bovine mitochondrial F1 (bMF1) and Paracoccus denitrificans F1 (PdF1), both of which exhibit higher maximum rotation rates than TF1. Systematic exploration of these sites identified four activity-enhancing hotspots, followed by focused hotspot exploration and machine-learning-assisted prioritization of combinatorial mutants. The best mutant, TF1({beta}Y313L/{beta}E332S), exhibited a 1.8-fold higher maximum rotation rate than TF1(WT) while retaining its functional thermostability. Interestingly, activity-enhancing substitutions were not limited to the residues conserved in both bMF1 and PdF1, indicating that the bMF1-PdF1 consensus substitutions effectively identify activity-enhancing hotspots rather than uniquely defining the optimal amino acid. Machine-learning-assisted exploration efficiently prioritized highly active mutants, although the predictive performance was limited by the relatively small training dataset and epistatic interactions among mutations. Kinetic and structural comparisons further provided mechanistic insights into the enhanced catalytic activity of the engineered mutant. Together, these results establish a practical strategy for engineering complex molecular motors by combining homolog-guided hotspot identification with focused hotspot exploration.

biophysics↗

bioRxiv2026

Characterization of Porcine Antibodies from Sequence Repertoire and Structural Data

Antibodies exhibit species-specific sequence and structural features that influence their antigen-recognition properties. Although several studies have investigated porcine antibodies, their repertoire and structural characteristics remain less well characterized than those of several other mammalian species. In this study, we analyzed public porcine heavy-chain repertoire sequencing data together with available antibody structural data to identify characteristic features of porcine antibodies. We found several residues enriched in porcine antibody framework regions, particularly at the base of heavy-chain complementarity-determining region 3 (CDR-H3). In particular, Arg101 and Glu123 were closely positioned in available structures and may influence CDR-H3 conformation at its base, whereas Pro120 may help constrain local backbone conformation. We also observed non-canonical cysteine usage in both framework region 1 and CDR-H3, which may contribute to structural diversity in the porcine repertoire. Finally, we evaluated the humanization potential of a porcine antibody using a human antibody language model and found that human-likeness increased after model-guided substitutions, although the resulting sequences did not exceed the T20 score threshold. Overall, these results indicate that porcine antibodies possess distinct sequence and structural features that may influence CDR-H3 properties and should be considered in future antibody analysis and engineering.

bioinformatics↗

bioRxiv2026

Evaluating Lightweight and Full Fine-Tuning Strategies Against Classical Machine Learning for Protein Function Prediction

Motivation Protein language models (PLMs) have emerged as powerful tools for sequence-based prediction of protein function, yet systematic benchmarks comparing frozen embeddings, fine-tuning strategies like Low-Rank Adaptation (LoRA) and classical machine learning (ML) remain limited. We benchmarked four ML strategies: ML using amino acid descriptors (SL-AAFeat), ML using frozen embeddings from 20 PLMs across various pooling strategies (SL-Embed), full model fine-tuning (FT-Full) and LoRA-based fine-tuning (FT-LoRA). Performance was evaluated on the in-house VHH phage display dataset (VHH) for binding affinity prediction and the TAPE fluorescence dataset (FLS and FLS10) for mutational effect prediction. Results Model performance depended strongly on the dataset and adaptation strategy. Max pooling consistently improved embedding-based models, while amino acid descriptors remained competitive under specific datasets and resource constraints. Fine-tuning generally provided the highest predictive performance, but the advantage is not universal. Hyperparameter optimization significantly enhanced FT-LoRA, enabling it to outperform FT-Full on the VHH dataset with less than 10% model parameter adaptation. In contrast, FT-Full achieved the best performance on FLS and FLS10. Several medium-sized PLMs performed comparably to larger models, highlighting favorable performance-efficiency trade-offs. Overall, this paper presents a thorough review of PLM utilization strategies and practical recommendations for selecting suitable strategies based on dataset characteristics and available computational resources. Availability The source code used in this manuscript is available in a Zenodo repository at https://doi.org/10.5281/zenodo.21466255.

bioinformatics↗

bioRxiv2026

Prior experience establishes a proactive cortical filter that blocks redundant associations

Prior experience allows selective learning by filtering out redundant cues, as exemplified by blocking, in which prior learning about one cue prevents a novel but redundant cue from being learned during subsequent compound conditioning. Top-down cortico-cortical pathways have been implicated in sensory filtering, but whether, how, and when they contribute to blocking remain unknown. Here we establish a head-fixed blocking paradigm in mice in which prior auditory cue-reward learning prevents a redundant somatosensory cue-reward association during subsequent compound conditioning. We found that a projection from anterior cingulate cortex (ACC) to primary somatosensory cortex (S1) facilitates blocking. Prior auditory learning establishes ACC-driven inhibitory control over S1 via parvalbumin neurons, before explicit somatosensory experience, thereby suppressing burst responses to the somatosensory cue, limiting network plasticity, and promoting blocking. These findings propose a proactive cortical filter that preconfigures inhibitory sensory bias to constrain future learning about redundant cues. Graphical abstract O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=200 SRC="FIGDIR/small/736748v1_ufig1.gif" ALT="Figure 1"> View larger version (63K): org.highwire.dtl.DTLVardef@6e1c4org.highwire.dtl.DTLVardef@1d01b49org.highwire.dtl.DTLVardef@1af67fdorg.highwire.dtl.DTLVardef@d42d8d_HPS_FORMAT_FIGEXP M_FIG C_FIG

neuroscience↗

bioRxiv2026

Exploring diverse routes to high-affinity-antibody variable domains through deep-sequencing-informed machine learning

The integration of in vitro selection, deep sequencing, and machine learning (ML) has recently been developed as a powerful strategy for discovering functional antibodies. However, how training data composition and ML search space design influence the identification of high-affinity variants remains unclear. Here, we aimed to optimize ML-integrated directed evolution for functional antibody discovery by selecting training data from deep sequencing analysis. By performing phage display selection using camelid heavy-chain antibodies (VHHs), we demonstrated that early-round data, retaining more binding-negative variants, can be superior for training models to identify high-performance VHHs. We also investigated a lead-independent ML search space design by focusing on conserved residues in final rounds, successfully identifying variants with higher affinities than those from lead-based maturation (KD = 7.9 nM). These findings demonstrate that training data selection and search space design are critical for successful ML-guided antibody engineering and provide diverse pathways for discovering high-affinity VHH variants.

bioengineering↗

bioRxiv2026

hnRNPA2B1 Modulates Early TIA1 Recruitment into Stress Granules through an RNA-Dependent Assembly Mechanism

Stress granules (SGs) are dynamic, membrane-less assemblies that form in the cytoplasm in response to cellular stress. The ordered recruitment of proteins into SGs is fundamental to condensate composition and function, yet the molecular determinants of this ordered client recruitment remain incompletely understood. Using proximity photo-crosslinking proteomics, we identified heterogeneous nuclear ribonucleoprotein A2B1 (hnRNPA2B1) as a TIA1-proximal protein preferentially enriched in SGs under arsenite stress. Knockdown of hnRNPA2B1 preferentially delayed TIA1 enrichment in G3BP1-marked SGs at 20 min without affecting G3BP1 or the overall SG-positive cell fraction, and this phenotype showed directional rescue upon re-expression. In vitro droplet reconstitution assays with purified proteins revealed that hnRNPA2B1 and RNA cooperatively increased TIA1 incorporation capacity into G3BP1 condensates, an effect not attributable to changes in droplet size. Kinetic fitting identified hnRNPA2B1 + RNA as uniquely increasing the plateau amplitude of TIA1 recruitment (Cohens d = 1.62 versus RNA-alone condition). Coarse-grained simulations support an inside-out assembly model in which hnRNPA2B1 stabilizes the condensate core through homotypic interactions while RNA-bound TIA1 accumulates at the periphery. Together, these findings identify hnRNPA2B1 as a capacity-determining modulator of early TIA1 recruitment and provide a framework for understanding ordered protein assembly within stress granules.

cell biology↗

bioRxiv2026

Fluid amyloid-β (Aβ) biomarkers reflect early β-sheet-rich Aβ deposition during the preclinical stage in Alzheimer's disease model 5XFAD mice

Early detection of disease progression using clinically-relevant biomarkers in animal models is important for mechanistic studies and for developing therapeutics in neurodegenerative diseases including Alzheimers disease (AD). The preclinical stage of AD, when amyloid-{beta} (A{beta}) starts to accumulate before cognitive decline, provides a critical window for disease modification. In humans, decreases in cerebrospinal fluid (CSF) A{beta}42 and the A{beta}42/A{beta}40 ratio in preclinical AD are considered to reflect the preferential sequestration of aggregation-prone A{beta}42 into {beta}-sheet-rich deposition in the brain, with corresponding changes being detectable in plasma. However, the extent to which these biomarker-pathology relationships are recapitulated in AD model mice remains incompletely defined. Here we show that CSF and plasma A{beta}42 and the A{beta}42/A{beta}40 ratio decline with age in parallel with the progression of {beta}-sheet-rich A{beta} deposition in preclinical 5XFAD mice, one of the most widely used AD mouse models, as assessed through monthly profiling of these biomarkers. Notably, the CSF A{beta}42/A{beta}40 ratio showed a negative correlation with {beta}-sheet-rich A{beta} deposition in the brain, whereas CSF A{beta}40 did not show a comparable association. In addition, the plasma A{beta}42/A{beta}40 ratio showed a positive correlation with the CSF A{beta}42/A{beta}40 ratio, suggesting that the plasma A{beta}42/A{beta}40 ratio may also reflect brain A{beta} deposition in this model. The strength of these correlations differed by sex, suggesting that sex-dependent differences in the A{beta} kinetics in this model may influence how closely fluid biomarkers reflect pathological progression. These findings support the potential utility of fluid A{beta} as a pathology-linked, translatable biomarker in preclinical 5XFAD mice. Highlights- Fluid A{beta} biomarkers are associated with early A{beta} deposition in preclinical 5XFAD mice. - The CSF A{beta}42/A{beta}40 ratio negatively correlates with {beta}-sheet-rich brain A{beta} deposition. - The plasma A{beta}42/A{beta}40 ratio positively correlates with the CSF A{beta}42/A{beta}40 ratio. - Monthly profiling defines fluid A{beta} biomarker dynamics in preclinical 5XFAD mice. - Sex differences may affect biomarker-pathology relationships in these mice.

neuroscience↗

bioRxiv2026

Prion-like transmission of human tau strains in the mouse brain

Most neurodegenerative diseases are believed to spread through the brain by prion-like mechanisms, where filamentous protein assemblies self-propagate by templated seeding 1. Distinct conformations of amyloid filaments are thought to provide the physical basis for the strains that lead to different diseases 2. However, a central pillar of the prion hypothesis, that strains retain their structural identity upon transmission, has not been demonstrated. Here we show that the injection of tau filaments from the brains of individuals with Alzheimers disease or corticobasal degeneration into the brains of wildtype mice leads to the seeded assembly of amyloid filaments made of mouse tau with the same structures as those of the seeds. Thereby, we establish that, like prion strains, tau filaments propagate through templated seeding, and that the mouse is a suitable model to study the molecular mechanisms by which distinct tau folds drive disease-specific pathology in the brain.

neuroscience↗

bioRxiv2026

Designing mRNA coding sequence via multimodal reverse translation language modeling with Pro2RNA

mRNA coding sequence design is a critical component in the development of mRNA vaccines, nucleic acid therapeutics, and heterologous gene expression systems. While large language models have recently been successfully applied to protein design and RNA modeling, designing optimal mRNA coding sequences for a given protein, particularly in a species-specific manner, remains a major challenge. Here, we present Pro2RNA, a multimodal reverse-translation language model that generates mRNA coding sequences from their corresponding protein sequences while explicitly conditioning on host organism taxonomy information. Pro2RNA integrates multiple pretrained language models across different modalities, including ESM2 for protein representation, SciBERT for taxonomy understanding, and a generative RNA language model for mRNA codon-level sequence generation. By training on mRNA-protein pairs from eukaryote and bacteria datasets respectively, Pro2RNA learns species-dependent genetic codes and codon usage patterns, enabling the generation of host-adapted and natural-like mRNA coding sequences. Across multiple benchmark evaluations, Pro2RNA matches or surpasses existing optimization methods, demonstrating its potential as a powerful and flexible framework for species-aware mRNA coding sequence design.

bioinformatics↗

bioRxiv2026

Lack of Period1 accelerates colorectal tumorigenesis in ApcMin/+ mice

The circadian clock coordinates physiology and behavior through [~]24-h rhythms, and disruption of core clock genes has been implicated in tumorigenesis. However, the impact of Per1, a major clock gene, on colorectal tumor development remains unclear. Here, we investigated how Per1 deletion influences intestinal tumorigenesis using the ApcMin/+ mouse model and ApcMin/+Per1-/-mice generated by crossing ApcMin/+ and Per1-/- lines (C57BL/6J background). Mice were maintained under controlled light-dark conditions, and we assessed survival, intestinal polyp burden, histopathology using Swiss-roll sections, {beta}-catenin protein abundance (immunofluorescence and western blotting), Ctnnb1 mRNA expression (RT-qPCR), and crypt proliferation (5-bromo-2-deoxyuridine (BrdU) immunohistochemistry). Per1 deletion did not significantly alter overall survival in ApcMin/+ mice but increased inter-individual variability. In contrast, polyp number was markedly increased by Per1 deletion, affecting both small (<2 mm) and large ([≥]2 mm) polyps across intestinal segments. Histology confirmed aberrant crypt foci and polyps in both ApcMin/+ genotypes. {beta}-Catenin protein levels in the whole intestine were significantly increased by Per1 deficiency and Apc mutation (two-way ANOVA), whereas Ctnnb1 mRNA was largely unchanged across regions. BrdU-based crypt proliferation was increased by the Apc mutation but not by Per1 deletion. These results indicate that Per1 loss exacerbates intestinal polyp formation and elevates {beta}-catenin predominantly through non-transcriptional mechanisms, supporting a tumor-suppressive role of Per1 in colorectal tumorigenesis.

cancer biology↗

bioRxiv2026

HTLV-1-Induced Neuroimmunome Correlates with Disease Progression and Severity

Human T-cell lymphotropic virus type 1 (HTLV-1) is a retrovirus that infects approximately 5-10 million people worldwide. While most individuals remain asymptomatic, a significant subset develops debilitating neuroinflammatory or malignant disorders, including adult T-cell leukemia/lymphoma (ATL) and HTLV-1-associated myelopathy/tropical spastic paraparesis (HAM/TSP). To unravel the systemic molecular mechanisms underlying HTLV-1 pathogenesis, we employed a multi-dimensional systems biology approach, integrating bulk transcriptomic data from total PBMCs (n = 200) with single-cell RNA sequencing (scRNA-seq) from 233,093 peripheral blood mononuclear cells (PBMCs). Our analysis revealed a consistent and clinically relevant neuroimmune signature within leukocytes, termed the neuroimmunome, comprising a set of differentially expressed genes shared across the nervous and immune systems. Through dimensionality reduction and machine learning techniques, such as PCA, gradient boosting, and MANOVA with bootstrapping, we identified potential biomarkers predictive of HTLV-1-driven leukemogenesis, which were subsequently validated across ATL, HAM/TSP, and asymptomatic cohorts via flow cytometry. Notably, expression levels of proteins such as ATF4 and SKIL were strongly correlated with proviral load, suggesting that sustained neuroimmune dysregulation may contribute to disease progression. These findings highlight a previously underappreciated neuroimmunological layer, redefining HTLV-1-associated disease as a condition deeply rooted in neuroimmune network disruption within leukocytes and offering potential novel targets.

molecular biology↗

bioRxiv2026

Vicious circle of amyloid and leptomeningeal macrophages evokes vascular dysfunction in CAA

The subarachnoid space contains leptomeningeal arteries and resident immune cells known as subarachnoid macrophages (SAMs). In Alzheimers disease (AD) and related cerebral amyloid angiopathy, amyloid-{beta} (A{beta}) frequently accumulates around the leptomeningeal arteries. Nevertheless, the local immune responses and their functional consequences remain poorly understood. Using longitudinal intravital imaging in a mouse model of AD, we tracked A{beta} deposition and identified a distinct SAM population that migrates to and phagocytoses these arterial A{beta} deposits. SAM recruitment correlated with vascular remodeling, including smooth muscle loss, aneurysm formation, and reduced cerebral perfusion. Transcriptomic profiling revealed a distinct SAM population in AD with upregulated CD39 expression. Pharmacological inhibition of CD39 attenuated arterial A{beta} deposition, identifying SAMs as a potential therapeutic target in AD.

neuroscience↗

bioRxiv2026

Dietary cholesterol reduces blood pressure and alters lipid profiles in stroke-prone spontaneously hypertensive rats

Although cholesterol (Chol) is widely recognized as a risk factor for cardiovascular disease, dietary Chol intake has been reported to extend the lifespan of stroke-prone spontaneously hypertensive rats (SHRSP). The mechanisms responsible for this paradoxical effect remain unclear. The present study examined changes in organ lipid profiles and associated molecular factors in SHRSP rats fed a Chol-enriched diet. Four-week-old male SHRSP/Izm rats were assigned to three groups and fed ad libitum for 12 weeks with either a control diet (Ctr), a diet supplemented with 1% w/w Chol (Chol), or a diet containing 1% w/w Chol plus 0.025% w/w lovastatin (Stt) to suppress endogenous Chol synthesis. Systolic blood pressure was measured before and after the feeding period, and tissues were collected for analyses of sterol content, fatty acid composition, prostaglandin E2 (PGE2) levels, and renal histopathology. Relative to the Ctr group, the Chol group exhibited a significant 9-10% reduction in systolic blood pressure. This reduction was accompanied by pronounced alterations in lipid profiles, including changes in phytosterol content and decreased arachidonic acid ratios in serum and kidney. There was a downward trend in hepatic PGE2 levels, and a similar tendency was observed in the kidney. Comparable changes in lipid profiles were observed in the Stt group. Histological analysis revealed modest attenuation of renal pathological features in Chol-fed rats. This study demonstrates for the first time that dietary Chol reduces renal phytosterol accumulation and suppresses the AA-PGE2 axis, changes that coincide with a 9-10% reduction in systolic blood pressure and attenuated glomerular inflammation. These integrated findings provide a mechanistic framework linking dietary Chol to the previously reported lifespan extension in this stroke-prone model. Although these changes may contribute to improved renal pathology, further studies are required to clarify causal relationships.

biochemistry↗

bioRxiv2025

Large generative mRNA language foundation model for efficient coding sequence generation and design with mRNA-GPT

mRNA design plays a central role in synthetic biology, nucleic acid therapeutics, and vaccine development. Although large language models are applied in many biological fields, generative language models for de novo mRNA design remains largely unexplored. Here, we introduce mRNA-GPT, a series of generative mRNA language models which for the first time covers the three domains of life as pretraining datasets. Based on a GPT-2 transformer architecture with 302 million parameters, we pre-trained three separate models on 19,676 bacterial, 4,688 eukaryotic, and 702 archaeal species, leveraging 80 million, 83 million, and 2 million mRNA coding sequences, respectively. Distinct clustering of mRNA coding sequence embeddings from animals, plants, and fungi in the pretrained mRNA-GPT-eukaryote indicates that the model captures organism-specific sequence features. Following unsupervised pre-training, we fine-tuned mRNA-GPT on a translation-efficiency dataset to generate high-performance mRNA sequence. Compared to the pretrained model, the fine-tuned mRNA-GPT produced mRNA sequences with significantly higher translation efficiency scores, demonstrating the ability of mRNA-GPT to capture sequence features underlying high translation efficiency. We further fine-tuned mRNA-GPT on datasets for mRNA stability and mRNA expression, where it likewise produced high-performance mRNA sequences. Our pretrained models are publicly available, enabling other researchers to adapt mRNA-GPT to specialized tasks such as tissue-specific mRNA expression or stability by fine-tuning on their own data. Together, our study demonstrates that generative mRNA language modeling as a promising approach for accelerating mRNA design across diverse biological fields.

bioinformatics↗

bioRxiv2025

High-Resolution Imaging of the Rhizosphere Using Resin-Embedded Cross-Sectional Polishing and Scanning Electron Microscopy for Biological Samples (bioCP-SEM)

Understanding microbial colonization in plant-associated environments requires high-resolution imaging of intact plant-microbe interfaces under natural soil conditions. However, conventional electron microscopy is limited by sample distortion and poor compatibility with heterogeneous substrates. Here, we present bioCP-SEM, a cross-sectional polishing scanning electron microscopy technique optimized for biological specimens inhabiting unsterilized soil. This method integrates resin embedding, diamond band saw sectioning, and mechanical polishing to enable high-quality imaging across large, complex samples. Using bioCP-SEM, we visualized bacterial clusters in a model plant root, and multilayered biofilm-like structures and fungal structures, including arbuscule-like forms, in a wild weed species. We further applied this method to intact moss collected from an urban environment, capturing spatially resolved interactions among biofilms, spores, protozoa, and annelids. Notably, microbial cells and subcellular structures, such as cilia, were preserved in situ. These results demonstrate the versatility of bioCP-SEM for mapping microbial architecture across diverse biological and environmental contexts.

microbiology↗

bioRxiv2025

Dietary soy shapes murine microbiota to consolidate the mucosal IgA response through T follicular helper cells

The commensal microbiota plays a crucial role in shaping mucosal immunity, particularly in the induction of T follicular helper (Tfh) cells and IgA production. Here, we demonstrate that dietary soy elicits a robust Tfh cell and IgA response in Peyers patches of weaning mice. Soy feeding promotes the expansion of two principal commensal bacterial species, Limosilactobacillus reuteri and Muribaculum intestinale. Mechanistically, L. reuteri provides cognate antigens for Tfh cell activation, while M. intestinale functions as an adjuvant by promoting IL-1{beta} production from myeloid cells. The resulting IgA exhibits polyreactivity and enhances protection against Salmonella infection. These findings highlight the specific interplay among dietary components, intestinal microbiota, and mucosal immunity, thereby establishing a diet-microbe-immune axis that shapes host defense in early life. This axis represents a promising therapeutic target for developing novel strategies to enhance resistance to enteric pathogens. Graphical abstract O_FIG O_LINKSMALLFIG WIDTH=170 HEIGHT=200 SRC="FIGDIR/small/669459v1_ufig1.gif" ALT="Figure 1"> View larger version (50K): org.highwire.dtl.DTLVardef@1ae8277org.highwire.dtl.DTLVardef@1f787ceorg.highwire.dtl.DTLVardef@1f51fd6org.highwire.dtl.DTLVardef@1fd3439_HPS_FORMAT_FIGEXP M_FIG C_FIG HighlightsO_LIDietary soy promotes T follicular helper (Tfh) cell and IgA responses in the Peyers patches of weaning mice. C_LIO_LISoy enhances the colonization of two key commensals, Limosilactobacillus reuteri and Muribaculum intestinale, in the small intestine. C_LIO_LIThese two bacteria synergize to drive mucosal immunity: L. reuteri provides cognate antigens, while M. intestinale provides an adjuvant signal by promoting IL-1{beta} production from myeloid cells. C_LIO_LISoy-induced IgA is polyreactive and protects against Salmonella infection. C_LI

immunology↗

bioRxiv2025

Data-efficient protein mutational effect prediction with weak supervision by molecular simulation and protein language models

Machine learning-based protein mutational effect prediction is widely used in protein engineering and pathogenicity prediction, but training data scarcity remains a major challenge due to high costs of experimental measurements. A previous study proposed data augmentation using computational estimates by molecular simulation. However, this approach has been limited to predicting mutational effects on thermostability. Here, we present a new data augmentation method that combines molecular simulation with zero-shot prediction computed by protein language models. These computational estimates serve as "weak" training data to supplement experimental training data. Our method dynamically adjusts the weight and inclusion of weak training data based on available experimental training data. This reduces potential negative impacts of weak training data while extending applicability to diverse protein properties such as binding affinity and enzymatic activity. Benchmark tests demonstrate that our method improves prediction accuracy particularly when experimental training data are scarce. These results indicate the capability of our approach to advance protein engineering and pathogenicity prediction in small data regimes.

bioinformatics↗