Biology subjects

Wang, D.

Publications and source records attributed to Wang, D..

At least 199 records · Page 11Linked to original sources

bioRxiv2025

OriGene: A Self-Evolving Virtual Disease Biologist Automating Therapeutic Target Discovery

Here, we present OriGene, a self-evolving multi-agent system that functions as a virtual disease biologist, systematically identifying original and mechanistically grounded therapeutic targets at scale. OriGenes architecture integrates over 600 specialized tools through a Model Context Protocol (MCP), enabling it to reason across diverse data modalities including genomics, protein networks, pharmacology, clinical records and literature evidence, to generate and prioritize target discovery hypotheses. We implemented a strategy combining a knowledge graph-based Tool RAG with an advanced agent selection mechanism to enable dynamic, context-aware tool deployment. Through a self-evolving framework, OriGene continuously integrates human and experimental feedback to iteratively refine its core thinking templates, tool composition, and analytical protocols, thereby enhancing both accuracy and adaptability over time. To comprehensively evaluate its performance, we established TRQA, an original benchmark comprising over 1,900 expert-level question-answer pairs spanning a wide range of diseases and target classes. OriGene consistently outperforms human experts, leading research agents, and state-of-the-art large language models in accuracy, recall, and robustness, particularly under conditions of data sparsity or noise. Critically, OriGene nominated previously underexplored therapeutic targets for liver (GPR160) and colorectal cancer (ARG2), which demonstrated significant anti-tumor activity in patient-derived organoid and tumor fragment models mirroring human clinical exposures. These findings demonstrate OriGenes potential as a scalable and adaptive platform for AI-driven discovery of mechanistically grounded therapeutic targets, offering a new paradigm to accelerate drug development.

bioinformatics↗

bioRxiv2025

Ultra-high field strength electroporation enables efficient DNA transformation and genome editing in nontuberculous mycobacteria

Efficient DNA delivery is essential for genetic manipulation of mycobacteria and for dissecting their physiology, pathogenesis, and drug resistance. Although electroporation enables transformation efficiencies exceeding 105 CFU per {micro}g DNA in Mycobacterium smegmatis and Mycobacterium tuberculosis, it remains highly inefficient in many non-tuberculous mycobacteria (NTM), including Mycobacterium abscessus. Here, we discovered that NTM such as M. abscessus exhibit exceptional tolerance to ultra-high electric field strengths, and that hypertonic preconditioning partially protects cells from electroporation-induced damage. Using ultra-high electric field strength (3kV/mm) electroporation, we achieved dramatic improvements in plasmid transformation efficiency--up to 106-fold in M. abscessus, 83-fold in Mycobacterium marinum, and 24-fold in Mycobacterium kansasii--compared to standard conditions (1.25 kV/ mm). Transformation efficiency was further influenced by the choice of selectable marker. Ultra-high field strength electroporation also markedly enhanced allelic exchange in M. abscessus expressing Che9c RecET recombinases, increasing recovery of gene deletion mutants by over 1,000-fold relative to conventional electroporation. In parallel, oligonucleotide-mediated recombineering for targeted point mutations produced nearly 10,000-fold more mutants under ultra-high field conditions. Together, these findings establish ultra-high field electroporation as a robust, broadly applicable platform for genetic engineering of NTMs. This method substantially enhances transformation efficiency and enables construction of advanced genetic tools--including expression libraries and CRISPRi knockdown libraries--in species that have historically resisted genetic manipulation. ImportanceInfections caused by non-tuberculous mycobacteria (NTM), including Mycobacterium abscessus, are increasing globally, yet genetic manipulation of these pathogens remains technically challenging due to inefficient DNA delivery and low gene editing success. The ultra-high electric field strength electroporation strategy described here overcomes these barriers, enabling dramatic improvements in both transformation and genome editing efficiency. This advance paves the way for high-throughput functional genomics in NTMs, including the construction of genome-wide knockout, CRISPRi knockdown, and expression libraries. Broad adoption of this approach will accelerate discovery of genetic determinants of virulence and drug resistance, facilitating the development of antimicrobials and vaccines.

microbiology↗

bioRxiv2025

Towards Unmanned Proteomics Data Generation: A Fully Automated Sample-to-Data System for Proteomic Experiments

The {pi}-HuB project requires proteomic measurements of millions of low-input samples that are challenging with current technology. Here, we introduce a pilot {pi}-HuB data factory that features a fully automated sample-to-data system to enable high-throughput digitization of human samples. Our platform, as a proof-of-concept towards unmanned proteomics, was demonstrated to show longitudinal robustness and applied for spatial proteomics and the profiling of plasma and cell lines, thereby serving as the workhorse to drive {pi}-HuB.

biochemistry↗

bioRxiv2025

A CRISPR/Cas9-based system using dual-sgRNAs for efficient gene deletion in Mycobacterium abscessus

The increasing global prevalence of Mycobacterium abscessus infections presents a significant clinical challenge due to the pathogens intrinsic resistance to multiple antibiotics and poor treatment outcomes. Despite the necessity of genetic tools for studying its physiology, pathogenesis, and drug resistance, efficient methods for large-fragment deletions remain underdeveloped. Here, we report a CRISPR/Cas9-based dual-sgRNA system employing Streptococcus thermophilus CRISPR1-Cas9 (Sth1Cas9), enabling efficient large-fragment knockout in M. abscessus with deletion efficiencies exceeding 90% at certain loci and spanning up to 16.7 kb. Furthermore, we systematically optimized the modular arrangement of genetic components in Cas9/dual-sgRNA expression plasmids and refined their construction workflow, achieving a significant reduction in cassette loss rates while enabling single-step plasmid assembly. Notably, deletion efficiency was position-dependent rather than correlated with target size, suggesting an influence of chromatin structure on editing outcomes. As the first CRISPR/Cas9-based platform capable of kilobase-scale deletions in M. abscessus, this system advances functional genomics studies and facilitates targeted investigations into virulence and antibiotic resistance mechanisms.

microbiology↗

bioRxiv2025

Complete Enzyme Clustering Enhances Coenzyme Q Biosynthesisvia Substrate Channeling

Metabolons - transient assemblies of sequential metabolic enzymes - facilitate the reactions of multi-step metabolic pathways, yet, how they mechanistically bolster metabolic flux remains unknown. Here, we investigate the molecular determinants of metabolon formation in coenzyme Q (CoQ) biosynthesis using coarse-grained molecular dynamics simulations and biochemical experiments. We show that the COQ metabolon forms at the critical region of a phase transition, where both metabolon clustering and metabolic flux exhibit coordinated sigmoidal responses to changes in protein-protein interaction strength. These complete metabolons enable substrate channeling between sequential enzymes, leading to a crucial enhancement of CoQ production efficiency. Selectively disrupting protein-protein interactions and randomly shuffling the interaction network demonstrate that protein-proximity rather than fine structure of the metabolon clusters is imperative for substrate channeling. Grounded in both experiment and simulation, these findings provide a framework for understanding the organization and function of metabolons across diverse metabolic pathways.

biophysics↗

bioRxiv2025

Biophysical modeling for accurate T cell specificity prediction of viral and tumor antigens

Accurate predictions of T cell receptor (TCR) specificity remain an important open problem in immunology, with broad implications for vaccine design, optimal immunotherapy, and improved management of autoimmune diseases. However, diversity in peptide antigens and TCR sequences at the level of individual patient repertoires remains a formidable computational challenge. Here, we develop a joint experimental and computational approach for predicting the antigen specificity of clinically-derived TCR sequences. Our model is trained on a combination of experimentally pre-identified and in silico-predicted TCR-pMHC structures using AlphaFold3. We apply our structural model in the clinical setting of hematopoietic stem cell transplant (HSCT) and demonstrate that our model is able to effectively discern the specificity of previously unseen donor and patient-derived TCR sequences against tumor associated and viral antigens. Model performance was further enhanced through the integration of sequence-based clustering and structurally diverse training templates. Our results highlight the predictive capabilities of structurally guided machine learning frameworks, trained on a minority test dataset, for antigen specificity prediction on unseen TCR sequences and their potential impact on a wide range of immunological applications.

immunology↗

bioRxiv2025

The RNA splicing factor PRPF8 is required for left-right organiser cilia function and determination of cardiac left-right asymmetry via regulation of Arl13b splicing

Cilia function in the left-right organizer (LRO) is critical for determining internal organ asymmetry in vertebrates. To further understand the genetics of left-right asymmetry, we isolated a mouse mutant with laterality defects, l11Jus27, from a random mutagenesis screen. l11Jus27 mutants carry a missense mutation in the pre-mRNA processing factor, Prpf8. cephaloph[o]nus (cph) mutant zebrafish, carrying a protein truncating mutation in prpf8, phenocopy the laterality defects of l11Jus27 mutants. Prpf8 mutant mouse and fish embryos have increased expression of an alternative transcript encoding the cilium-associated protein, ARL13B, that lacks exon 9. In zebrafish, over-expression of the arl13b transcript lacking exon 9 perturbed cilium formation and caused laterality defects. The shorter ARL13B protein isoform lacked interactions with intraflagellar transport proteins. Our data suggest that PRPF8 plays a prominent role in LRO cilia by through the regulation of alternative splicing of ARL13B, thus uncovering a new mechanism for cilia-linked developmental defects.

developmental biology↗

bioRxiv2025

Evolutionary responses to historic drought across the range of scarlet monkeyflower

Adaptive evolution is a key means for populations to persist under environmental change, yet whether populations across a species range can adapt quickly enough to keep pace with climate change remains unknown. The breeders equation predicts the evolutionary change in a trait from one generation to the next as the product of the selection differential and the narrow-sense heritability in that trait. Incorporating these aspects of the breeders equation, we performed a resurrection study with the scarlet monkeyflower (Mimulus cardinalis) to evaluate whether traits associated with drought adaptation have evolved in populations across a species range in response to extreme drought. We compared trait and fitness differences of pre-drought ancestors and post-drought descendants from six populations transplanted into three latitudinally-arrayed common gardens and quantified phenotypic selection and trait heritabilities. The strength, direction, and mode of selection varied among traits and gardens. Trait heritabilities were relatively low, and did not differ dramatically among populations or gardens. Overall, instances of evolutionary responses between ancestors and descendants were few and small in magnitude, but the magnitude of these evolutionary differences varied among gardens. Together, these results suggest that the expression of genetic variation, and thus traits, depend on the environment, and that environmental variability in field settings may mask the genetic variation that is often detected in greenhouse environments.

evolutionary biology↗

bioRxiv2025

Mechanism-driven screening of membrane-targeting and pore-forming antimicrobial peptides

The rise of antibiotic resistance has generated an urgent demand for the discovery of new antimicrobial peptides (AMPs), prompting the development of various screening strategies. However, the specific function mechanisms of AMPs are often overlooked during the screening and optimization processes. In this study, we introduce a mechanism-driven screening approach that employs machine learning-based computational models to identify peptide sequences that target membranes and form pores. This approach explicitly considers critical factors such as structural features, membrane ainity, and the ability of peptides to oligomerize. Our method was applied to the metaproteomes of poison frogs, African clawed frogs, and human skin, followed by experimental validation. Seven peptides were successfully screened, each demonstrating antimicrobial activity with minimal hemolysis and cytotoxicity. These peptides exhibited membrane disruption capabilities in liposome leakage assays, with three showing broad-spectrum antimicrobial activity. Furthermore, single-molecule experiments indicated that these peptides can oligomerize on membranes, while electrophysiological measurements detected pore formation, confirming the effectiveness of our screening strategy. Therefore, our screening approach can effectively identify AMP sequences that act through membrane-targeting and poreforming mechanisms, offering a promising, mechanism-driven strategy for the discovery of new antimicrobial agents to combat antibiotic resistance.

biophysics↗

bioRxiv2025

A Universal 6iL/E4 Culture System for Deriving and Maintaining Embryonic Stem Cells Across Mammalian Species

The derivation of authentic embryonic stem cells (ESCs) from diverse mammalian species offers valuable opportunities for advancing regenerative medicine, studying developmental biology, and enabling species conservation. Here, we report the development of a robust, serum-free culture system, termed 6iL/E4 that enables the derivation and long-term self-renewal of ESCs from multiple mammalian species, including mouse, rat, bovine, rabbit, and human. Using systematic signaling pathway analysis, we identified key regulators--including GSK3, STAT3, PDGFR, BRAF, and LATS--critical for ESC maintenance across species. Additionally, inducible expression of KLF2 and NANOG enhances the naive pluripotency and chimeric potential of bovine ESCs. The E4 medium also supports stable ESC growth while minimizing lineage bias. These findings reveal conserved principles underlying ESC self-renewal across divergent mammalian species and provide a universal platform for cross-species stem cell research, disease modeling, and biotechnology applications. In BriefWang et al. developed 6iL/E4, a serum-free system sustaining ESCs from mouse, rat, bovine, rabbit, and human. These findings reveal conserved fundamental mechanisms governing ESC self-renewal across diverse mammalian species. HighlightsO_LIDeveloped 6iL/E4 system for ESC derivation across five mammalian species. C_LIO_LIPDGFR signaling inhibition as critical for ESC derivation across species. C_LIO_LIE4 medium improves ESC maintenance and avoids neural bias of traditional N2B27. C_LIO_LIInducible KLF2/NANOG enhances naive pluripotency and chimera formation in bovine. C_LI

cell biology↗

bioRxiv2025

Selective GSK3α Inhibition Promotes Self-Renewal Across Different Stem Cell States

Pan-GSK3/{beta} inhibition promotes stem cell self-renewal through activation of WNT/{beta}-catenin signaling, but its broad effects complicate the precise control of stem cell states. Here, we show that selective inhibition of GSK3 with BRD0705 supports the long-term self-renewal of mouse embryonic stem cells (ESCs), epiblast stem cells (EpiSCs), and neural stem cells (NSCs), independent of {beta}-catenin signaling. When combined with the tankyrase inhibitor IWR1, BRD0705 broadly supports the maintenance of diverse pluripotent stem cell states, including ESCs, EpiSCs, and formative pluripotent stem cells. This BRD0705/IWR1 cocktail enables stable co-culture of naive ESCs and primed EpiSCs while preserving their distinct molecular and functional identities. Single-cell transcriptomics, epigenomic profiling, and functional assays confirm sustained lineage-specific features across stem cell types. These findings demonstrate that selective GSK3 inhibition enhances stemness by buffering against differentiation cues and promoting intrinsic self-renewal capacity. This work identifies GSK3 as a key regulator of self-renewal across distinct stem cell states and establishes a versatile culture system with broad applications. In BriefWang et al. demonstrate that selective GSK3 inhibition with BRD0705 supports self-renewal of pluripotent and neural stem cells. Combined with IWR1, it enables long-term co-culture of naive and primed stem cells while preserving their distinct molecular and functional identities. Graphical abstract O_FIG O_LINKSMALLFIG WIDTH=181 HEIGHT=200 SRC="FIGDIR/small/653860v1_ufig1.gif" ALT="Figure 1"> View larger version (36K): org.highwire.dtl.DTLVardef@4e371eorg.highwire.dtl.DTLVardef@104b287org.highwire.dtl.DTLVardef@164bd96org.highwire.dtl.DTLVardef@daf1ba_HPS_FORMAT_FIGEXP M_FIG C_FIG HighlightsO_LIGSK3 inhibition by BRD0705 promotes self-renewal of ESCs, EpiSCs, and NSCs C_LIO_LIBRD0705/IWR1 enables long-term co-culture of ESCs and EpiSCs C_LIO_LICo-cultured ESCs and EpiSCs retain distinct naive or primed identities C_LIO_LIBRD0705 preserves stem cell states independently of {beta}-catenin signaling C_LI

cell biology↗

bioRxiv2025

Sequential Amino Acid Mutagenesis-Driven De Novo Evolution of Adenine Deaminases Enables Efficient In Vivo Base Editing in Primate

Base editing allows for the precise modification of genetic information, providing new avenues for treating diseases1. The adenine base editor ABE8e is currently the most efficient and widely used tool for adenine base editing. ABE8e was developed through multiple rounds of directed evolution of Escherichia coli tRNA adenine deaminase, including phage-assisted continuous evolution (PACE)2,3. While PACE is highly effective, it is a complex system that poses challenges for implementation4. Despite its high efficiency, ABE8e is associated with limitations such as relatively higher bystander editing effects and elevated off-target activity5, which need to be addressed to further enhance its precision and safety. Here, we developed a novel method for the de novo discovery of evolved ABE components, particularly adenine deaminases. This process involves identifying candidate proteins through AI-based structural prediction and clustering, followed by the enhancement of deaminase editing activity through screening libraries created by sequential amino acid saturation mutagenesis. This evolutionary strategy simplifies the approach by employing saturation mutagenesis libraries tailored to specific segments, thereby enabling exploration of an expanded sequence space and increasing the likelihood of discovering adenine deaminases with superior capabilities. The newly developed hpABE5.20 here demonstrates a more refined editing window, reduced DNA off-target effects that are both sgRNA-dependent and -independent, and minimized RNA off-target activity, while maintaining robust editing efficiency relative to ABE8e. Furthermore, hpABE5.20 has been successfully applied for precise and effective therapeutic adenine base editing in cellular disease models, humanized mice, and non-human primates.

bioengineering↗

bioRxiv2025

AF3Score: A Score-Only Adaptation of AlphaFold3 for Biomolecular Structure Evaluation

Scoring biomolecular complexes remains central to structural modeling efforts. Recent studies suggest that AlphaFold (AF) - a revolutionary deep learning model for biomolecular structure prediction - has implicitly learned an approximate biophysical energy function. While many researchers highly rely on AF-derived scores for structure evaluation, existing AlphaFold2-based implementations require iterative refinement of the input structure, leading to biased scoring. To address this limitation, we adapted AlphaFold3 into a score-only model, AF3Score, by directly feeding input coordinates into the confidence head while bypassing the diffusion-based structure module. AF3Score demonstrates robust performance in structural quality assessment across diverse systems including monomeric proteins, protein-protein complexes, de novo designed binders, and fold-switching proteins. In benchmarking designed binder screening, AF3Score outperformed state-of-the-art methods for 8 out of 10 targets. Moreover, combining AF3Score with AlphaFold2-derived methods significantly improved the enrichment of experimentally validated binders, increasing the success rate from 15.2% to 31.6%. Additionally, AF3Score effectively identified stable conformations in fold-switching proteins, whereas AlphaFold predominantly predicted only the dominant fold. These findings highlight the broad applicability of AF3Score, from high-throughput screening in de novo binder design to filtering docking-generated poses and molecular dynamics (MD) trajectories.

bioinformatics↗

bioRxiv2025

A 3D Modeling Framework for Quantifying Variation in Soybean Root Structure Architecture

Root system architecture (RSA) underpins plant access to water and nutrients, making its characterization critical for improving crop performance in environments with limited soil fertility. However, current methods for quantifying root features face several challenges. They may rely on 2D images that suffer from occlusion, use expensive sensing technologies like X-ray computed tomography, or depend on 3D modeling approaches with assumptions about branching that make them difficult to generalize. To address these challenges, we introduce an open-source Python framework for quantifying RSA samples from 3D point clouds generated from low-cost photogrammetry. Critically, this method incorporates no assumptions about taxon-specific branching orientation, making it both well-suited for modeling naturally grown annual dicots such as soybean and generalizable across species. Using field-grown soybean as a test case, we demonstrate the utility of this framework to extract biologically meaningful 3D features of divergent root systems sampled across developmental stages and soil environments, and enable new analyses not possible with 2D approaches, such as modeling metabolic scaling relationships. Results indicate that, in our soybean samples, while certain individual features like taproot tortuosity are potentially influenced by the soil environment, and while roots in sandy loam exhibited greater feature plasticity, fundamental scaling properties remain consistent. By combining low-cost photogrammetry with 3D reconstruction of root systems from point clouds, this approach provides the plant science community with new opportunities for more comprehensive root studies.

plant biology↗

bioRxiv2025

Loading causes molecular damage in fibrin fibers

Blood clotting is the bodys natural reaction in wound healing and is also the cause of many pathologies. Fibrin - the main protein in the clotting process provides clots mechanical strength by forming a scaffold of complex fibrin fibers. Fibrin fibers exhibit high extensibility and primarily elastic properties under static loading, which differ from in vivo dynamic forces. In many biological materials, the mechanical response changes under repeated loading/unloading (cyclic loading). Using lateral force microscopy, we show fibrin fibers possess viscoelastic behavior and experience irreversible damage under cyclic loading. Cross-linking results in a more rigid structure with permanent damage occurring mostly at larger strains, which is corroborated by computational modeling of fibrin extension using a hyperelastic model. Molecular spectroscopy analysis with broadband coherent anti-Stokes Raman scattering spectroscopy in addition to molecular dynamic simulations allow identification of the source of damage, the unfolding pattern, and inter and intramolecular changes in fibrin. The results show partial recovery of proteins secondary and tertiary structures under load, providing deeper understanding of fibrins unique behavior in wound healing or pathologies like stroke and embolism.

biophysics↗

bioRxiv2025

Function of the auditory cortex characterized by its intrinsic dynamic coactivation patterns estimated in individuals

Determining the functional organization of the auditory cortex (AC) has been difficult with conventional task-based approaches due to the broad responsiveness of auditory subregions to various acoustic properties. Moreover, most studies have investigated functional organization of AC with static methods, although brain has shown to be organized into dynamic networks. Here, we investigated dynamically varying coactivation patterns of the local networks in the auditory cortex (AC) determined from 7T fMRI data with a novel individualized network-based algorithm. An eight-pattern solution was selected for closer examination based on its high reproducibility of the occurrence rates (r=0.86) and the spatial topography (r=0.79) between sessions. Dynamic AC patterns successfully captured interindividual variability, as indicated by significantly higher variability between than within individuals for the AC pattern occurrence rates and spatial topographies. The coactivation patterns shared similarities between resting-state and auditory-task data, as indicated by the group-level similarity of 0.84 and individual-level similarity of 0.71 in the spatial topographies. Furthermore, the occurrence rates of AC patterns identified in the task data, using pattern templates derived from resting-state data, correlated with specific task contrast regressors. Our results indicate that the AC function can be characterized by a set of dynamically varying coactivation patterns that are consistently observed during resting state and auditory stimulation, and that get synchronized with auditory input. These findings enhance our understanding of the relationship between spontaneous and stimulus-driven activity in the AC and support the development of more time-efficient paradigms for studying its functional organization.

neuroscience↗

bioRxiv2025

Stretchable, hair-compatible, and long-term stable wearable EEG system

Electroencephalography (EEG) is a cornerstone in both neuroscience research and clinical diagnostics. However, conventional EEG monitoring faces hardware limitations, particularly its adaptability and stability. Headsets either require complicated wiring or do not have enough stretchability and wearability to comply with the diverse head anthropometry and hair conditions of the user population. Additionally, there is an inherent tradeoff between wet and dry electrodes in capturing high-fidelity signals from hair-covered scalp regions while ensuring continuous and long-term recording quality. Here, we present a Mesh-integrated, Stretchable, and Hair-compatible EEG system engineered to overcome these limitations. By incorporating a kirigami-inspired mesh design and stretchable eutectic Gallium-Indium interconnects, MindStretcH adapts to various head sizes and allows for easy wearing and removal. Moreover, its uniquely designed porous, conical, soft 3D-printed mold, embedded with hydrogel electrodes, effectively penetrates hair layers to deliver low impedance and sustained signal integrity with minimal discomfort. We validate MindStretcH through offline and online EEG-based brain-computer interface tasks over a month, demonstrating its exceptional stability in continuous monitoring and dynamic applications. These results mark a promising advance toward non-invasive neural interfaces in both clinical and everyday use.

bioengineering↗

bioRxiv2025

Enhancing Structure-aware Protein Language Models with Efficient Fine-tuning for Various Protein Prediction Tasks

Proteins are crucial in a wide range of biological and engineering processes. Large protein language models (PLMs) can significantly advance our understanding and engineering of proteins. However, the effectiveness of PLMs in prediction and design is largely based on the representations derived from protein sequences. Without incorporating the three-dimensional structures of proteins, PLMs would overlook crucial aspects of how proteins interact with other molecules, thereby limiting their predictive accuracy. To address this issue, we present S-PLM, a 3D structure-aware PLM that employs multi-view contrastive learning to align protein sequences with their 3D structures in a unified latent space. Previously, we utilized a contact map-based approach to encode structural information, applying the Swin-Transformer to contact maps derived from AlphaFold-predicted protein structures. This work introduces a new approach that leverages a Geometric Vector Perceptron (GVP) model to process 3D coordinates and obtain structural embeddings. We focus on the application of structure-aware models for protein-related tasks by utilizing efficient fine-tuning methods to achieve optimal performance without significant computational costs. Our results show that S-PLM outperforms sequence-only PLMs across all protein clustering and classification tasks, achieving performance on par with state-of-the-art methods that require both sequence and structure inputs. S-PLM and its tuning tools are available at https://github.com/duolinwang/S-PLM/.

bioinformatics↗