Biology subjects

Li, S. Z.

Publications and source records attributed to Li, S. Z..

16 recordsLinked to original sources

bioRxiv2026

Osteopontin Mediates Uterine Artery Dysfunction in Hypertensive Pregnancy

BackgroundSuccessful pregnancy requires substantial maternal cardiovascular adaptation, including expansion and remodelling of the uterine arteries to support increased nutrient flow to the fetus. Hypertension is associated with impaired uterine artery remodelling and increased risk of fetal growth restriction and other adverse outcomes, yet the mechanisms driving vascular dysfunction in hypertensive pregnancy remain incompletely understood. Osteopontin, a matricellular protein, is implicated in vascular pathology in hypertension, positioning it as a candidate mediator of impaired uterine artery adaptation during hypertensive pregnancy. MethodsUterine artery remodelling was compared between pregnant normotensive Wistar-Kyoto (WKY) rats and spontaneously hypertensive stroke-prone rats (SHRSP), a model of chronic hypertension. To determine the role of osteopontin in this process, an osteopontin-deficient SHRSP strain was characterized. Uterine artery blood flow was assessed by Doppler ultrasound, arterial structure was evaluated by histology, and molecular differences were identified by RNA sequencing. Fetal weight and length were measured at mid and late gestation. ResultsCompared with WKY, SHRSP fetuses were smaller, and uterine arteries exhibited inward hypertrophic remodelling, characterized by increased wall thickness, reduced lumen area, and elevated resistance index. SHRSP uterine arteries also showed increased expression of inflammatory and vascular pathology-associated genes, including osteopontin. In osteopontin-deficient SHRSP, uterine arteries had larger lumen areas, decreased resistance index, and reduced expression of inflammation-associated genes. Fetal growth was also improved in osteopontin-deficient SHRSP pregnancies. ConclusionsThese findings identify osteopontin as a contributor to impaired uterine artery adaptation and suggest that reduced osteopontin may improve vascular remodelling and fetal growth in hypertensive pregnancies.

developmental biology↗

bioRxiv2026

Multimodal physical evidence uncovers interpretable gene regulatory networks for perturbation prediction

Gene regulatory networks govern cell fate transitions through dynamic causal mechanisms1. Since exhaustively mapping this vast perturbation space experimentally is prohibitive2, scalable computational models are essential. Yet, current frameworks fall short because they infer statistical co-expression rather than physical mechanisms, remain blind to non-canonical regulators lacking classical DNA-binding motifs, and fail to generalize across unseen perturbation factors or cell lines. Here we show that a multimodal biophysical framework, VitaGRN, overcomes these barriers by constructing a biophysical regulatory scaffold from multimodal evidence and propagating interactions to capture non-canonical regulators. By leveraging structurally aligned protein embeddings, VitaGRN predicts zero-shot perturbation responses and uncovers non-canonical translational control programs. Notably, VitaGRN demonstrates robust generalization across unseen factors, cell lines, and developmental transitions. Ultimately, VitaGRN generates a confidence-calibrated virtual perturbation atlas spanning over a thousand factors. This resource reframes gene regulatory networks from static correlation graphs into dynamically generalizable and mechanistically trans-parent models, streamlining wet-lab candidate prioritization.

bioinformatics↗

bioRxiv2025

Expanding the RNA Virus Universe by Scalable Structure-Guided Discovery

The discovery of RNA viruses from metatranscriptomic data remains challenging due to their extreme sequence divergence and frequent lack of conserved motifs. We present Rider, a lightweight two-stage framework that couples fast, structure-informed sequence screening with targeted structural validation. Stage 1 uses a compact 35M-parameter protein language model to prioritize RdRp-like fragments at whole-sample scale, achieving over 44x higher end-to-end screening throughput on commodity hardware. Stage 2 applies structure prediction and Foldseek-based alignment against a dedicated RdRp structure resource ([~]200k ESMFold-predicted structures), providing orthogonal evidence for remote homologs. Applied to >10,000 metatranscriptomes spanning marine, freshwater, soil and host-associated microbiomes, Rider matches or outperforms leading tools (e.g., LucaProt, PalmScan) and additionally recovers divergent and truncated sequences. Multiple orthogonal indicators, including structure consistency and low DNA read mapping to corresponding contigs, support genuine RNA origin. In a human IBD cohort, Rider agrees with state-of-the-art calls for clinically relevant RNA viruses while extending discovery to divergent lineages. Rider turns structure-guided homology search into a practical, scalable pipeline for RNA virome discovery. HighlightO_LIA two-stage framework enables structure-guided RNA virus discovery at sample scale, achieving up to 44-fold higher throughput on standard computing hardware. C_LIO_LIThe method matches or surpasses LucaProt and PalmScan across >10,000 metatranscriptomes from diverse environments, while recovering RdRp fragments missed by existing tools. C_LIO_LIStructural validation using [~]200,000 ESMFold-predicted RdRp models and Foldseek alignment supports the detection of remote homologs with high confidence. C_LIO_LIOrthogonal evidence, including low DNA read mapping, strand-specific expression, and ORF metrics, confirms RNA origin and reduces false positives.. C_LIO_LIOpen-source code and an openly released RdRp structure database enable scalable, reproducible RNA virome discovery in environmental and clinical settings. C_LI

bioinformatics↗

bioRxiv2024

Bridging the Gap between Database Search and De Novo Peptide Sequencing with SearchNovo

AO_SCPLOWBSTRACTC_SCPLOWAccurate protein identification from mass spectrometry (MS) data is fundamental to unraveling the complex roles of proteins in biological systems, with peptide sequencing being a pivotal step in this process. The two main paradigms for peptide sequencing are database search, which matches experimental spectra with peptide sequences from databases, and de novo sequencing, which infers peptide sequences directly from MS without relying on pre-constructed database. Although database search methods are highly accurate, they are limited by their inability to identify novel, modified, or mutated peptides absent from the database. In contrast, de novo sequencing is adept at discovering novel peptides but often struggles with missing peaks issue, further leading to lower precision. We introduce SearchNovo, a novel framework that synergistically integrates the strengths of database search and de novo sequencing to enhance peptide sequencing. SearchNovo employs an efficient search mechanism to retrieve the most similar peptide spectrum match (PSM) from a database for each query spectrum, followed by a fusion module that utilizes the reference peptide sequence to guide the generation of the target sequence. Furthermore, we observed that dissimilar (noisy) reference peptides negatively affect model performance. To mitigate this, we constructed pseudo reference PSMs to minimize their impact. Comprehensive evaluations on multiple datasets reveal that SearchNovo significantly outperforms state-of-the-art models. Also, analysis indicates that many retrieved spectra contain missing peaks absent in the query spectra, and the retrieved reference peptides often share common fragments with the target peptides. These are key elements in the recipe for SearchNovos success. The code for reproducing the results are available in the supplementary materials.

bioinformatics↗

bioRxiv2024

FoldToken4: Consistent & Hierarchical Fold Language

Creating protein structure language has attracted increasing attention in unifing the modality of protein sequence and structure. While recent works, such as FoldToken1&2&3 have made great progress in this direction, the relationship between languages created by different models at different scales is still unclear. Moreover, models at multiple scales (different code space size, like 25, 26, [···], 212) need to be trained separately, leading to redundant efforts. We raise the question: Could a single model create multiscale fold languages? In this paper, we propose FoldToken4 to learn the consistent and hierarchical of multiscale fold languages. By introducing multiscale code adapters and token mixing techniques, FoldToken4 can generate multiscale languages from the same model, and discover the hierarchical token-mapping relationships across scales. To the best of our knowledge, FoldToken4 is the first effort to learn multi-scale token consistency and hierarchy in VQ research; Also, it should be more novel in protein structure language learning.

molecular biology↗

bioRxiv2024

FoldToken3: Fold Structures Worth 256 Words or Less

Protein structure tokenization has attracted increasing attention in both protein representation learning and generation. While recent work, like FoldToken2 and ESM3, has achieved good reconstruction performance, the compressoin ratio is still limited. In this work, we propose FoldToken3, a novel protein structure tokenization method that can compress protein structures into 256 tokens or less and ensure the reconstruction quality comparable to FoldToken2. To the best of our knowledge, FoldToken3 is the most efficient, light-weight, and compression-friendly protein structure tokenization method. And it will benifit a wide range of protein structure-related tasks, such as protein structure alignment, generation, and representation learning. The work is still in progress and the code will be available upon acceptance.

molecular biology↗

bioRxiv2024

FoldToken2: Learning compact, invariant and generative protein structure language

The equivariant nature of 3D coordinates has posed long term challenges in protein structure representation learning, alignment, and generation. Can we create a compact and invariant language that equivalently represents protein structures? Towards this goal, we propose FoldToken2 to transfer equivariant structures into discrete tokens, while maintaining the recoverability of the original structures. From FoldToken1 to FoldToken2, we improve three key components: (1) invariant structure encoder, (2) vector-quantized compressor, and (3) equivariant structure decoder. We evaluate FoldToken2 on the protein structure reconstruction task and show that it outperforms previous FoldToken1 by 20% in TMScore and 81% in RMSD. FoldToken2 is likely the first method that works well for both single-chain and multi-chain protein structure quantization. We believe that FoldToken2 will inspire further improvement in protein structure representation, alignment, and generation tasks. Online example is available at Colab.

molecular biology↗

bioRxiv2024

GeoAB: Towards Realistic Antibody Design and Reliable Affinity Maturation

Increasing works for antibody design are emerging to generate sequences and structures in Complementarity Determining Regions (CDRs), but problems still exist. We focus on two of them: (i) authenticity of the generated structure and (ii) rationality of the affinity maturation, and propose GEOAB as a solution. In specific, GeoABDesigner generates CDR structures with realistic internal geometries, composed of a generative geometry initializer (Geo-Initializer) and a position refiner (Geo-Refiner); GeoAB-Optimizer achieves affinity maturation by accurately predicting both the mutation effects and structures of mutant antibodies with the same network architecture as Geo-Refiner. Experiments show that GEOAB achieves state-of-the-art performance in CDR co-design and mutation effect predictions, and fulfills the discussed tasks effectively.

bioinformatics↗

bioRxiv2024

Effectiveness and efficiency: label-aware hierarchical subgraph learning for protein-protein interaction (laruGL-PPI)

Protein-protein interactions (PPIs) are crucial in various biological processes and their study has significant implications for drug development and disease diagnosis. Existing deep learning methods for PPIs prediction, including graph neural networks (GNNs), have been widely employed as the solutions, while they usually suffer from performance degradation under complex real-world scenarios. We claim that the topological shortcut is one of the key problems contributing negatively to the performance, according to our analysis. By modeling the PPIs as a graph with protein as nodes and interactions as edge types, the prevailing models tend to learn the pattern of nodes degrees rather than intrinsic sequence-structure profiles, leading to the problem termed topological shortcut. In addition, with the emergence of high-throughput experimental methods such as mass spectrometry and protein chip technology, the amount of available PPI data is exploding. The huge data growth leads to intensive computational costs and challenges computing devices, causing infeasibility in practice. To address the discussed problems, we propose a label-aware hierarchical subgraph learning method (laruGL-PPI) that can effectively infer PPIs while being both interpretable and generalizable. Specifically, we introduced edge-based subgraph sampling to effectively alleviate the problems of topological shortcuts and high computing costs. Besides, the inner-outer connections of PPIs are modeled as a hierarchical graph, together with the dependencies between interaction types constructed by a label graph. Extensive experiments on PPI datasets of different scales demonstrate that laruGL-PPI outperforms state-of-the-art PPI prediction methods, particularly in the testing of unseen proteins. Also, our model can recognize crucial sites of proteins, such as surface sites for binding and active sites for catalysis.

bioinformatics↗

bioRxiv2024

PPFlow: Target-aware Peptide Design with Torsional Flow Matching

Therapeutic peptides have proven to have great pharmaceutical value and potential in recent decades. However, methods of AI-assisted peptide drug discovery are not fully explored. To fill the gap, we propose a target-aware peptide design method called PPFO_SCPLOWLOWC_SCPLOW, based on conditional flow matching on torus manifolds, to model the internal geometries of torsion angles for the peptide structure design. Besides, we establish a protein-peptide binding dataset named PPBench2024 to fill the void of massive data for the task of structure-based peptide drug design and to allow the training of deep learning methods. Extensive experiments show that PPFO_SCPLOWLOWC_SCPLOW reaches state-of-the-art performance in tasks of peptide drug generation and optimization in comparison with baseline models, and can be generalized to other tasks including docking and side-chain packing.

molecular biology↗

bioRxiv2023

Learning Complete Protein Representation by Deep Coupling of Sequence and Structure

Learning effective representations is crucial for understanding proteins and their biological functions. Recent advancements in language models and graph neural networks have enabled protein models to leverage primary or tertiary structure information to learn representations. However, the lack of practical methods to deeply co-model the relationships between protein sequences and structures has led to suboptimal embeddings. In this work, we propose CoupleNet, a network that couples protein sequence and structure to obtain informative protein representations. CoupleNet incorporates multiple levels of features in proteins, including the residue identities and positions for sequences, as well as geometric representations for tertiary structures. We construct two types of graphs to model the extracted sequential features and structural geometries, achieving completeness on these graphs, respectively, and perform convolution on nodes and edges simultaneously to obtain superior embeddings. Experimental results on a range of tasks, such as protein fold classification and function prediction, demonstrate that our proposed model outperforms the state-of-the-art methods by large margins.

bioinformatics↗

bioRxiv2023

Multiple conformational states assembly of multidomain proteins using evolutionary algorithm based on structural analogues and sequential homologues

With the breakthrough of AlphaFold2, nearly all single-domain protein structures can be built at experimental resolution. However, accurate modelling of full-chain structures of multidomain proteins, particularly all relevant conformations for those with multiple states remain challenging. In this study, we develop a multidomain protein assembly method, M-SADA, for assembling multiple conformational states. In M-SADA, a multiple population-based evolutionary algorithm is proposed to sample multiple conformational states under the guidance of multiple energy functions constructed by combining homologous and analogous templates with inter-domain distances predicted by deep learning. On a developed benchmark dataset containing 72 multidomain proteins with multiple conformational states, the performance of M-SADA is significantly better than that of AlphaFold2 on multiple conformational states modelling, where 29/72 (40.3%) of proteins can be assembled with a TM-score >0.90 for highly distinct conformational states with M-SADA while AlphaFold2 does so in only 2/72 (2.8%) of proteins. Furthermore, M-SADA is tested on a developed benchmark dataset containing 296 multidomain proteins with single conformational state, and results show that the average TM-score of M-SADA on the best models is 0.913, which is 5.2% higher than that of AlphaFold2 models (0.868).

bioinformatics↗

bioRxiv2022

Research on protein structure prediction and folding based on novel remote homologs recognition

Recognition of remote homologous structures is a necessary module in AlphaFold2 and is also essential for the exploration of protein folding pathways. Here, we developed a new method, PAthreader, which identifies remote homologous structures based on the three-track alignment of distance profiles and structure profiles originated from PDB and AlphaFold DB by deep learning. Based on the identified templates, we further enhanced state-of-the-art modelling method and explored protein folding pathways based on the residue frequency distribution of homologs and the secondary structure. The results show that the average accuracy of templates identified by PAthreader is 11.6% higher than those of HHsearch on 551 nonredundant proteins. In terms of structure modelling, PAthreader improves the performance of AlphaFold2 and ranks first in CAMEO blind test for the last three months. Furthermore, we explored protein folding pathways for 37 proteins. The results are almost consistent with biological experiments for 7 proteins, and the remaining 30 human proteins have yet to be verified by biological experiments, revealing that folding information can be exploited from remote homologous structures.

bioinformatics↗

bioRxiv2022

Structure-Preserving and Batch-Correcting Visualization Using Deep Manifold Transformation for Single-cell RNA-Seq Profiles

Dimensionality reduction and visualization play an important role in biological data analysis, such as data interpretation of single-cell RNA sequence (scRNA-seq). It is desired to have a visualization method that can not only be applicable to various application scenarios, including cell clustering and trajectory inference, but also satisfy a variety of technical requirements, especially the ability to preserve inherent structure of data and handle with batch effects. However, no existing methods can accommodate these requirements in a unified framework. In this paper, we propose a general visualization method, deep visualization (DV), that possesses the ability to preserve inherent structure of data and handle batch effects and is applicable to a variety of datasets from different application domains and dataset scales. The method embeds a given dataset into a 2- or 3-dimensional visualization space, with either a Euclidean or hyperbolic metric depending on a specified task type with type static (at a time point) or dynamic (at a sequence of time points) scRNA-seq data, respectively. Specifically, DV learns a structure graph to describe the relationships between data samples, transforms the data into visualization space while preserving the geometric structure of the data and correcting batch effects in an end-to-end manner. The experimental results on nine datasets in complex tissue from human patients or animal development demonstrate the competitiveness of DV in discovering complex cellular relations, uncovering temporal trajectories, and addressing complex batch factors. We also provide a preliminary attempt to pre-train a DV model for visualization of new incoming data.

bioinformatics↗

bioRxiv2022

Towards Effective and Generalizable Fine-tuning for Pre-trained Molecular Graph Models

Graph Neural Networks (GNNs) and Transformer have emerged as dominant tools for AI-driven drug discovery. Many state-of-the-art methods first pre-train GNNs or the hybrid of GNNs and Transformer on a large molecular database and then fine-tune on downstream tasks. However, different from other domains such as computer vision (CV) or natural language processing (NLP), getting labels for molecular data of downstream tasks often requires resource-intensive wet-lab experiments. Besides, the pre-trained models are often of extremely high complexity with huge parameters. These often cause the fine-tuned model to over-fit the training data of downstream tasks and significantly deteriorate the performance. To alleviate these critical yet under-explored issues, we propose two straightforward yet effective strategies to attain better generalization performance: 1. MolAug, which enriches the molecular datasets of down-stream tasks with chemical homologies and enantiomers; 2. WordReg, which controls the complexity of the pre-trained models with a smoothness-inducing regularization built on dropout. Extensive experiments demonstrate that our proposed strategies achieve notable and consistent improvements over vanilla fine-tuning and yield multiple state-of-the-art results. Also, these strategies are model-agnostic and readily pluggable into fine-tuning of various pre-trained molecular graph models. We will release the code and the fine-tuned models.

bioinformatics↗

bioRxiv2020

Phenotype Prediction using a Tensor Representation and Deep Learning from Data Independent Acquisition Mass Spectrometry

A novel approach for phenotype prediction is developed for mass spectrometric data. First, the data-independent acquisition (DIA) mass spectrometric data is converted into a novel file format called "DIA tensor" (DIAT) which contains all the peptide precursors and fragments information and can be used for convenient DIA visualization. The DIAT format is fed directly into a deep neural network to predict phenotypes without the need to identify peptides or proteins. We applied this strategy to a collection of 102 hepatocellular carcinoma samples and achieved an accuracy of 96.8% in classifying malignant from benign samples. We further applied refined model to 492 samples of thyroid nodules to predict thyroid cancer; and achieved a predictive accuracy of 91.7% in an independent cohort of 216 test samples. In conclusion, DIA tensor enables facile 2D visualization of DIA proteomics data as well as being a new approach for phenotype prediction directly from DIA-MS data.

bioinformatics↗