Biology subjects

Wen, B.

Publications and source records attributed to Wen, B..

At least 19 recordsLinked to original sources

bioRxiv2026

Improving real-time search through deep learning-enhanced fragment ion indexing

The fragment ion indexed database searching approach enables highly efficient tandem mass spectrometry database searches, including real-time search workflows. However, the index can require tens of gigabytes of memory, often exceeding the free memory available on instrument acquisition computers running the search. This restricts the size of the searchable database, the variable modification space, and the number of searches that can run concurrently. To address this problem, we developed an index-pruning pipeline within the Comet database search tool using Carafe, a deep learning tool that predicts fragment ion intensities based on peptide sequences. In our approach, only fragment ion positions that are predicted to be observable are written in the index, eliminating entries that are unlikely to contribute to spectrum detection. We evaluated this approach on two search spaces in three index configurations to demonstrate that masking significantly reduced the number of fragment ion index entries and, consequently, peak search memory use. Masked searches were also faster than unmasked searches and identified a comparable number of peptide-spectrum matches. Because all prediction and masking steps are performed before searching, their cost is amortized across repeated, future searches of the same database and modification configuration, introducing no per-spectrum search overhead. The complete implementation is open source with an Apache license.

bioinformatics↗

bioRxiv2026

Rac1 and CHK1 Converge on Abi-1 to Regulate DNA Repair Dependency and Treatment Response in Human Cancers

Radiation therapy (RT) resistance remains a major clinical challenge, yet biomarkers guiding precision radiosensitization are lacking. We previously demonstrated that Rac1 promotes RT resistance in glioblastoma (GBM) by inducing Abi-1-S323 dephosphorylation and enhancing non-homologous end joining (NHEJ). Here, we identify Abi-1-S323 as a key regulator of DNA repair states and a determinant of therapeutic efficacy in human cancers. Clinically, loss of Abi-1-S323 phosphorylation was associated with poor outcomes in patients with RT-treated GBM. Bioinformatic analyses revealed that non-small cell lung cancer (NSCLC) and head and neck cancer (HNC) are among the cancers with frequent RAC1 amplification, suggesting that these tumor types may have increased Rac1-Abi-1 signaling activity. Loss of Abi-1-S323 phosphorylation also predicted poor outcomes in patients with RT-treated HNC. Consistent with these clinical observations, high Rac1 activity and low Abi-1-S323 phosphorylation were associated with enhanced DNA double-strand break repair and radioresistance in NSCLC and HNC models, whereas genetic or pharmacological inhibition of this signaling impaired DNA repair and radiosensitized tumors in vitro and in vivo. Mechanistically, we identified CHK1 as a kinase that phosphorylates Abi-1 at S323 and defines an alternative homologous recombination (HR)-dependent repair state. Tumors with high Rac1-Abi-1 signaling exhibited elevated NHEJ capacity and were selectively radiosensitized by Rac1 inhibition, whereas tumors with low Rac1-Abi-1 signaling displayed high CHK1 activity, preferentially relied on HR, and were selectively radiosensitized by CHK1 inhibition. These findings establish Abi-1-S323 as a biomarker defining therapeutically distinct DNA repair states and provide a framework for precision radiosensitization.

cancer biology↗

bioRxiv2026

CasanovoGUI: a cross-platform desktop application for deep learning-based de novo peptide sequencing with Casanovo

De novo peptide sequencing detects peptides directly from tandem mass spectra without a protein sequence database, and deep learning has substantially advanced its performance. Casanovo, one such widely used model, is distributed as a Python command-line program. Consequently, installation, GPU and dependency configuration, and manual parameterization can be challenging for many bench scientists and are a recurring source of errors. Interpreting and validating the resulting predictions poses a further challenge. We present CasanovoGUI, an open-source Java-based desktop application that makes all of Casanovo's main analysis functions available through a point-and-click interface on Windows, macOS, and Linux. On first use, CasanovoGUI automatically installs a private Python environment and Casanovo with a GPU-matched build, requiring no prior software setup. The GUI provides access to Casanovo's analysis functions and configuration parameters, streams live progress, and integrates results interpretation: annotated spectra with per-residue confidence scores in the PDV viewer, and mismatch-tolerant mapping of de novo peptides back to a reference proteome. CasanovoGUI is available at https://github.com/Noble-Lab/CasanovoGUI.

bioinformatics↗

bioRxiv2026

CryoDiff: An uncertainty-aware diffusion model for Cryo-EM map enhancement

Cryogenic electron microscopy (cryo-EM) enables high-resolution structural determination of large macromolecular complexes. However, the interpretability of cryo-EM maps is often hindered by substantial background noise and signal attenuation, which obscure structural details. Although existing post-processing methods can partially mitigate these artifacts, they typically suffer from over-smoothing and lack reliable confidence estimation. Here, we present CryoDiff, an uncertainty-aware diffusion model for cryo-EM map enhancement. CryoDiff employs a multi-step diffusion process to progressively denoise and restore high-resolution structural features. Importantly, CryoDiff incorporates a voxel-wise confidence metric derived from Monte Carlo sampling. It unifies map enhancement and voxel-level uncertainty estimation within a diffusion-based generative framework, representing the first approach to achieve such joint modeling for cryo-EM map enhancement. In comprehensive experiments, CryoDiff markedly out-performs existing methods in both map-model correlation and map interpretability, improving the average FSC0.5 metric by 0.356 [A] over state-of-the-art approaches. When applied to de novo model building with ModelAngelo, CryoDiff further increases model completeness by 5.5%, exceeding the gains achieved by competing method.

bioinformatics↗

bioRxiv2026

Revisiting resonance-excitation collision-induced dissociation for data-independent acquisition

Data-independent acquisition (DIA) proteomics relies almost exclusively on beam-type collision-induced dissociation (HCD) because its short activation time supports fast acquisition rates required for DIA. However, HCD requires charge state calibration and is therefore imperfect for mixed-charge DIA isolation windows. Resonance-excitation collision-induced dissociation (reCID) offers a promising alternative to HCD for DIA because the ion activation is effectively independent of charge state. Historically, reCIDs longer activation time has been considered too slow for DIA. Here, we revisit reCID for DIA proteomics using a modified Orbitrap Tribrid Apex MultiOmics mass spectrometer (Apex) that recovers acquisition-matrix overhead as additional ion injection time, enabling reCID acquisition rates comparable to HCD. Using matched acquisition rate settings of tryptic HeLa digests, reCID achieved precursor and protein detections similar to HCD when used with Carafe fine-tuned, fragmentation-matched spectral libraries. Library fine-tuning improved reCID precursor detections more than HCD detections, 24% versus 5%, indicating that HCD-trained prediction models are suboptimal for reCID spectra. ReCID also maintained peptide-level quantitative performance, including ions measured per peptide, precision, and accuracy. Across seven NCI cancer cell lines and pooled mixtures, protein abundance rankings were highly conserved between the Apex reCID and HCD methods, and also across Orbitrap Astral Zoom platforms. These results support reCID as a practical fragmentation mode for DIA proteomics.

biochemistry↗

bioRxiv2026

Prioritizing peptides for targeted mass spectrometry experiments using deep learning

One critical step in any targeted mass spectrometry experiment is selecting, from each protein of interest, a small number of peptides that respond well in the mass spectrometer and can serve as reliable proxies for protein quantification. Existing methods select target peptides either by relying on prior empirical measurements, limiting their applicability to previously observed peptides, or using machine learning to predict peptide behavior from sequence alone. However, current machine learning tools suffer from various limitations, including using detectability as an indirect proxy for intensity, relying on small training sets, or ignoring the precursor charge state. In this study, we introduce Bromo, a transformer-based deep learning model that ranks peptide precursors from a given protein by their relative response, taking charge state into account. Trained on millions of annotated peptide pairs derived from large-scale, publicly available data-independent acquisition mass spectrometry data, Bromo consistently outperforms existing sequence-based methods across diverse, independent datasets. Furthermore, we show that fine-tuning Bromo on experiment-specific data can account for differences in sample preparation, sample matrix, and instrument platform, all of which influence which peptides serve as optimal targets. This adaptability makes Bromo a practical tool for selecting target peptides for selected reaction monitoring and parallel reaction monitoring assay development across a wide range of experimental conditions.

bioinformatics↗

bioRxiv2026

Virtual multiplex staining of the pancreatic islets across type 1 diabetes progression using a Schroedinger bridge

Classical hematoxylin and eosin (H&E) staining enables review of tissue morphology but lacks information regarding the molecular state of cells. Immunohistochemical (IHC) techniques label specific proteins in tissue, allowing differentiation of relevant structures that may go undetectable in H&E. However, the IHC process is complex, expensive, and time-consuming, especially for multiplex IHC (mIHC) limiting its use in large cohorts. Stain conversion of H&E to IHC using generative artificial intelligence models such as generative adversarial networks (GANs) represent one solution to this problem. However, GANs are unstable during out of distribution sampling and are prone to hallucinations or mode collapse, limiting their accuracy in challenging image conversion tasks. To address this, the field has recently turned to diffusion models. Here, we introduce Schrodinger-bridge for Multiplex ImmunoLabel Estimation (SMILE). Unlike conventional diffusion models that map from source to target through an intermediate Gaussian noise, Schrodinger-bridge diffusion models skip this step and have been shown to better preserve structures during image translation. To test the performance of SMILE, we generated a large cohort of high-fidelity H&E-mIHC image pairs from pancreatic organ donors, targeting insulin, glucagon, and CD3. Our dataset well-sampled across type-1 diabetes status, pancreas anatomical location, age, and sex. Using this cohort, we demonstrate the superiority of SMILE compared to GANs via a comprehensive evaluation framework incorporating texture, distribution, and antibody-specific metrics, as well as blinded pathologist reviews. We further confirmed the ability of SMILE to generate accurate mIHC images from H&Es generated at an external site, to perform whole slide image conversion, and to generate realistic three-dimensional maps of the pancreatic islets in non-diabetic, auto-antibody positive, and type-1 diabetic donor tissue. Finally, we performed stain conversion of paired H&E to HER2 and Ki67 images in breast cancer, confirming the superiority of SMILE in diverse stain conversion applications. Collectively, this framework provides a scalable pipeline for high-throughput proteomic inference from archival H&Es, providing transformative potential for pancreatic research and digital pathology.

bioinformatics↗

bioRxiv2026

Carafe2 enables high quality in silico spectral library generation for timsTOF data-independent acquisition proteomics

Data-independent acquisition (DIA) proteomics enables reproducible and systematic peptide detection and quantification, and trapped ion mobility spectrometry (TIMS) on the timsTOF platform further improves DIA by synchronizing ion mobility separation with quadrupole precursor sampling. Analyzing the highly multiplexed spectra generated by DIA typically relies on spectral libraries, and fully leveraging the additional ion mobility dimension requires these libraries to include accurate retention time, fragment ion intensity, and ion mobility annotations. Existing in silico spectral library generation tools either lack ion mobility support entirely or rely on models trained on data-dependent acquisition (DDA) data, that can introduce a mismatch that may not capture unique experiment-specific biases when applied to each respective timsTOF dataset. Carafe is a software tool that uses deep learning models to generate high-quality, experiment-specific in silico libraries by training directly on DIA data. In this study, we extend Carafe to generate libraries for timsTOF DIA data, which involves fine-tuning retention time (RT), fragment ion intensity, and ion mobility prediction models using timsTOF DIA data. Carafe2 operates directly on native timsTOF raw data (Bruker .d directories) without the need for data conversion. We demonstrate the performance of Carafe2 across a wide range of DIA applications, including global proteome, phosphoproteome, and plasma proteome datasets. Comparing Carafe2 fine-tuned RT, fragment ion intensity, and ion mobility prediction models with pretrained DDA models, we find that Carafe2 models outperform pretrained models on a variety of DIA datasets. We then demonstrate the utility of in silico libraries generated by Carafe2 for peptide detection on several different types of timsTOF DIA datasets by comparing with the libraries generated with DDA-trained AlphaPeptDeep models, DIA-NN built-in models, and empirical spectral libraries generated from DDA experiments.

bioinformatics↗

bioRxiv2026

Comprehensive bioinformatic analysis reveals novel potential diagnostic biomarkers associated with monocytes in osteoporosis

Osteoporosis affects millions of women globally. In this study, we applied bioinformatics methods to screen for novel diagnostic biomarkers of osteoporosis in women using the GSE62402 and GSE56814 datasets. PCSK5, ZNF225, and H1FX were used to construct a diagnostic model. ROC, calibration, and decision curve analyses were performed to assess the diagnostic performance on the training (GSE56814) and external (GSE56815) datasets. The expression level of model genes was validated in GEO datasets. Furthermore, five transcription factors (ETS1, NOTCH1, MAZ, ERG, and FLI1) were identified as common upstream regulators of model genes. PCSK5, ZNF225, and H1FX serve as novel diagnostic biomarkers, providing new insights into the pathogenesis of and treatment strategies for osteoporosis in women.

genetics↗

bioRxiv2026

Geometric structure of features underlies human VTC object recognition.

The ventral temporal cortex (VTC) plays a crucial role in human object recognition. Within VTC neural space, object-specific domains align with domain-general features, e.g. face- and scene-domain align with the feature distribution of animacy, generating a hierarchical knowledge structure for categorization. However, the neural process of integrating information from VTC to distinguish different objects remains unclear. Here, we employed a combination of ANN modeling, functional MRI, and MEG to investigate how VTC features affect object manifold separability. The representational geometry analysis shows that domain-general features in VTC form a unique structure, different from ANN, to assist object classification. Moreover, VTC dynamically adjusts the geometrical relationship of these features during object recognition, influencing the geometrical properties of object manifolds and, consequently, their separability. These findings advance our understanding of the neural computation involved in VTC object recognition, revealing how downstream neurons can flexibly access category information in diverse recognition tasks.

neuroscience↗

bioRxiv2025

Puget predicts gene expression across cell types using sequence and 3D chromatin organization data

Gene expression is governed by both linear DNA sequence and three-dimensional (3D) chromatin architecture. Most gene expression prediction models rely on sequence alone, thereby failing to capture structural context and to generalize to unseen cell types. We present Puget, a deep learning model that predicts cell type-specific gene expression from sequence and Hi-C data, which captures 3D chromatin organization. Puget pairs pretrained sequence and Hi-C encoders with a lightweight transformer decoder. Using paired Hi-C/RNA-seq from 36 human and 4 mouse biosamples, we evaluate Pugets ability to generalize to held-out genes, held-out biosamples, and from human to mouse. Relative to a sequence-only baseline, Puget improves cross-biosample Pearson correlation by up to 25% on highly variable genes in training biosamples and, unlike the sequence-only model, generalizes to held-out biosamples and across species. In addition, in silico perturbation experiments show that Puget can prioritize experimentally validated enhancer-gene pairs. Together, these results highlight a generalizable approach for modeling gene expression from sequence and 3D chromatin organization.

bioinformatics↗

bioRxiv2025

Albumin-STING Nanocomplex Reprograms HSPCs to Antitumor Neutrophils with Enhanced MHC I Antigen Presentation for Cancer Immunotherapy

Tumor-associated immunosuppressive neutrophils, termed myeloid-derived suppressor cells (MDSCs), compromise cancer immunotherapy. They can be pathologically programmed as early the hematopoietic stem and progenitor cell (HSPC) stage by suppressing interferon signaling. Reprogramming HSPCs toward antitumor neutrophils through the stimulator of interferon genes (STING) activation offers a promising therapeutic strategy. Here, we demonstrate that an albumin-STING nanoagonist (Nano ZSA-51D) reprograms HSPCs to generate antitumor neutrophils, enhancing MHC I-mediated CD8 T cell immunity. Nano ZSA-51D activates STING-interferon signaling in HSPCs, promoting their expansion and differentiation toward granulocyte-monocyte progenitors via STING-NF-{kappa}B-IL-6 signaling. It further reprograms neutrophils into CD14ICAM-1+ subset through STING-NF-{kappa}B-TNF- signaling, enhancing tumor infiltration. These neutrophils upregulate interferon signaling and MHC I antigen presentation, boosting tumor-specific CD8 T cell responses. Adoptive transfer of Nano ZSA-51D-reprogrammed neutrophils with -PD1 therapy achieves complete colon tumor remission. Our findings provide a novel strategy to reprogram HSPCs toward antitumor neutrophils and highlight the potential of early interventions at HSPC stage to rewire neutrophil fate for cancer immunotherapy.

immunology↗

bioRxiv2025

bacNeo: a computational toolkit for bacteria-derived neoantigen identification

Intratumour microbiome can restore anti-tumour immunity with bacterial peptides potentially serving as neoantigens, while robust in-silico tools for bacterial neoantigen identification remain lacking. We developed bacterial neoantigen (bacNeo), a multi-omics based computational toolkit that contains three modules to classify bacterial taxonomy, identify host HLA alleles, and predict bacterial neoantigens. bacNeo could enable personalised neoantigen discovery to advance therapeutic strategies across diverse cancer types.

bioinformatics↗

bioRxiv2025

Improvements to Casanovo, a deep learning de novo peptidesequencer

Casanovo is a state-of-the-art deep learning model for de novo peptide sequencing from mass spectrometry proteomics data. Here we report on a series of enhancements to Casanovo, aimed at improving the interpretability of the scores assigned to predicted peptides, generalizing the software for use in database search, speeding up training and prediction runtimes, and providing workflows and visualization tools to facilitate adoption of Casanovo and interpretation of its results. Our goal is to make Casanovo accurate and easy to use for applications such as metaproteomics, antibody sequencing, immunopeptidomics, and discovery of novel peptide sequences in standard proteomics analyses. Casanovo is available as open source at https://github.com/Noble-Lab/casanovo.

molecular biology↗

bioRxiv2025

Analysis of isobaric quantitative proteomic data using TMT-Integrator and FragPipe computational platform

Isobaric mass tags, such as iTRAQ and TMT, are widely utilized for peptide and protein quantification in multiplex quantitative proteomics. We present TMT-Integrator, a bioinformatics tool for processing quantitation results from TMT and iTRAQ experiments, offering integrative reports at the gene, protein, peptide, and post-translational modification site levels. We demonstrate the versatility of TMT-Integrator using five publicly available TMT datasets: an E. coli dataset with 13 spike-in proteins, the clear cell renal cell carcinoma (ccRCC) whole proteome and phosphopeptide-enriched datasets from the Clinical Proteomic Tumor Analysis Consortium, and two human cell lysate datasets showcasing the latest advances with the Astral instrument and TMT 35-plex reagents. Integrated into the widely used FragPipe computational platform (https://fragpipe.nesvilab.org/), TMT-Integrator is a core component of TMT and iTRAQ data analysis workflows. We evaluate the FragPipe/TMT-Integrator analysis pipelines performance against MaxQuant and Proteome Discoverer with multiple benchmarks, facilitated by the bioinformatics tool OmicsEV. Our results show that FragPipe/TMT-Integrator quantifies more proteins in the E. coli and ccRCC whole proteome datasets, identifies more phosphorylated sites in the ccRCC phosphoproteome dataset, and delivers overall more robust quantification performance compared to other tools.

bioinformatics↗

bioRxiv2025

Label-free quantification in the Crux toolkit

Ultimately, most tandem mass spectrometry (MS/MS) proteomics experiments aim to not just detect but also quantify the proteins in a given complex sample. Here, we describe an extension to the Crux MS/MS analysis toolkit to enable label-free quantification of peptides. We demonstrate that Cruxs new quantification command, which is modeled after the algorithms implemented in the widely used FlashLFQ software, is both efficient and accurate. In particular, we achieve a 1.9-fold speedup while reducing the memory usage by 26%. The new crux-lfq command is available in Crux v4.3.

bioinformatics↗

bioRxiv2024

Patient-Derived Xenografts of Triple-Negative Breast Cancer Enable Deconvolution and Prediction of Chemotherapy Responses

Chemotherapy regimens for triple-negative breast cancer (TNBC) combine agents without knowing which agents drive response. Consequently, predictors derived from multi-agent regimens cannot be assumed to generalize to individual drugs or other regimens, motivating development of treatment-matched predictors. Here, we used TNBC patient-derived xenografts (PDXs) treated with carboplatin, docetaxel, or the combination to deconvolute drug-specific responses and identify associated molecular features. Combination treatment rarely improved upon the best single agent, with enhanced responses in only 13% of PDXs and antagonism in a comparable fraction. Proteogenomic analyses identified high cytokeratin-5 (KRT5) as a general marker of chemotherapy responsiveness and KRT5 immunohistochemistry discriminated responsive PDXs (AUROC, 0.83). To train treatment-specific predictors, we integrated these data with independent PDX and clinical cohorts with responses assessed after anthracycline-free platinum, taxane, or platinum-taxane therapy, ensuring response corresponded to the modeled treatment. Four feature selection strategies yielded largely nonoverlapping biomarker panels converging on treatment-relevant pathways. On independent test data, RNA-based predictors of complete response (CR) to platinum-based (carboplatin or cisplatin) and taxane-based (docetaxel or paclitaxel) chemotherapy achieved AUROCs of 0.80 and 0.86, respectively. For platinum-taxane regimens (carboplatin plus docetaxel or paclitaxel), proteomic-guided feature selection generated a 10-biomarker, protein- informed RNA predictor of pathologic complete response (pCR) that outperformed RNA-only feature selection and achieved an AUROC of 0.85 in an independent clinical cohort, while retaining practicality as an RNA-based assay. Treatment-matched integration of multi-omic PDX and clinical datasets provides a framework for chemotherapy-specific predictors with clinically relevant performance, supporting biomarker-guided precision selection and treatment optimization for patients with TNBC. Statement of significanceIntegration of multi-omic data from patient-derived xenografts with treatment-matched clinical cohorts yielded three retrospectively validated predictors of platinum, taxane, and platinum+taxane response, enabling biomarker-guided chemotherapy selection for patients with triple-negative breast cancer.

cancer biology↗

bioRxiv2024

Carafe enables high quality in silico spectral library generation for data-independent acquisition proteomics

Data-independent acquisition (DIA)-based mass spectrometry is becoming an increasingly popular mass spectrometry acquisition strategy for carrying out quantitative proteomics experiments. Most of the popular DIA search engines make use of in silico generated spectral libraries. However, the generation of high-quality spectral libraries for DIA data analysis remains a challenge, particularly because most such libraries are generated directly from data-dependent acquisition (DDA) data or are from in silico prediction using models trained on DDA data. In this study, we developed Carafe, a tool that generates high-quality experiment-specific in silico spectral libraries by training deep learning models directly on DIA data. We demonstrate the performance of Carafe on a wide range of DIA datasets, where we observe improved fragment ion intensity prediction and peptide detection relative to existing pretrained DDA models. To make Carafe more accessible to the community, we have integrated Carafe into the widely used Skyline tool.

bioinformatics↗