Biology subjects

Xu, D.

Publications and source records attributed to Xu, D..

At least 55 records · Page 3Linked to original sources

bioRxiv2025

Adapting Co-Folding Models for Structure-Based Protein-Protein Docking Through Flow Matching

Co-folding models like AlphaFold have revolutionized protein complex structure prediction, yet their reliance on multiple sequence alignments (MSAs) limits their applicability on challenging targets such as antibody-antigen complexes. An alternative approach, structure-based protein-protein docking, predicts the bound complex structure from the unbound monomer structures without requiring MSAs. In this work, we propose a novel method to adapt co-folding models for structure-based protein-protein docking by replacing their template module with a docking module, followed by training end-to-end with a flow-matching objective. We apply our method to AlphaFold-Multimer (AF-M) using the OpenFold implementation and transform it into a generative docking model, which we name AF2Dock. We evaluate AF2Dock and various baseline methods on the PINDER-AF2 benchmark and an antibody/nanobody test set. When using non-holo inputs, AF2Dock shows competitive performance compared to other structure-based docking methods and, in the case of nanobody complexes, outperforms all other docking methods tested here. Although AF2Dock underperforms co-folding AF-M and AF3 in success rates when using non-holo inputs, it produces orthogonal predictions and successfully identifies correct structures for targets where co-folding models fail. Ablation studies confirm that full-parameter fine-tuning of the AF-M components is critical for performance and reveal that, surprisingly, the inclusion of ESM embeddings can hinder success rates in certain cases such as nanobody complexes. The code is available at https://github.com/Graylab/AF2Dock.

biophysics↗

bioRxiv2025

Can We Extract Physics-like Energies from Generative Protein Diffusion Models?

Diffusion models have emerged as the state-of-the-art method in generative artificial intelligence (AI) and have shown great success in image synthesis, video generation, molecular design, and protein structure prediction. For biophysical problems, such as protein folding and association, a fundamental question in diffusion-based methods is how their learned functions correspond to thermodynamics. In this paper, we study diffusion models through the lens of theoretical biophysics, analyzing their underlying formulation of potentials and exploring their applications in scoring protein interactions. We develop simple theories rooted in statistical physics that relate thermodynamic potentials to the negative log of the probability of observing a system in a particular state. We include dimensional analysis of diffusion model equations and a table mapping AI and physics jargon. We then test a diffusion models ability to capture learned energies as negative log-likelihood values, -log p0(x0), by integrating over the diffusion-generated path or a probability flow path. We test these integrals on a simple 1D Gaussian mixture diffusion model and a protein-docking diffusion model, DFMDock. In the 1D case, we find that integration over both diffusion and flow paths can accurately recover ground truth probabilities. When we extract the learned docking energies for cases where DFMDock succeeds, we observe energy funnels with the minimum energy near the experimental docked structure, like those we observe with Rosetta, an empirically tuned physics-based biomolecular modeling suite. The learned energy performs comparably or outperforms Rosetta interface energy in 9 out of 25 cases at ranking the correctness of docked poses. These data show that we can extract a relevant learned energy function from a diffusion model and compare it to physical energy functions.

biophysics↗

bioRxiv2025

Proline transporters balance the salicylic acid-mediated trade-off between regeneration and immunity in plants

A robust immune response and regenerative capacity are both essential for survival after injury. In plants, salicylic acid (SA) is essential for activating immunity but simultaneously suppresses regenerative capacity. The mechanisms coordinating the trade-off between immunity and regeneration remain poorly understood. Here, we identify proline transporters as key regulators of this balance. Mutations in two wound-induced proline transporters, ProT2 and ProT3, rescued exogenous proline-induced suppression of de novo root regeneration (DNRR) and enhanced DNRR. ProTs are required for SA-mediated suppression of regeneration, without affecting SA-dependent defense responses. Mechanistically, a ProT3-CPK1 complex modulates the dynamics of wound-induced reactive oxygen species (ROS), sustaining a ROS level that restricts DNRR after wounding. Notably, pharmacological inhibition of proline transport rescued SA-mediated suppression of regeneration and enhanced regeneration across multiple plant species. These findings establish proline transporter as a regulatory hub integrating stress-induced proline metabolism, SA signaling, and ROS homeostasis to balance immunity and regeneration, and highlight chemical inhibition of proline transport as a strategy to improve crop regeneration under biotic stresses without compromising disease resistance.

plant biology↗

bioRxiv2025

Privacy-Preserving Genomic Data Sharing via Hybrid AES/ECC and Blockchain-Based Dynamic Consent

Genomic data are growing exponentially, creating opportunities for discovery but also raising significant concerns regarding privacy, security, and governance. We present a framework that encrypts genome-scale files using AES-256-GCM and protects keys per recipient via Curve25519, anchors only content identifiers and encrypted keys on a blockchain smart contract, and stores ciphertext off-chain in IPFS. The design enables dynamic, auditable consent through on-chain access-control lists (grant/revoke with immutable events) while keeping raw genomes off-chain. A working prototype on the Ethereum Sepolia testnet demonstrates end-to-end registration, retrieval, and revocation. Experiments on the human reference genome (hg38) and synthetic sequences demonstrate near-AES performance for multi-gigabyte files and strong diffusion, as per the avalanche criterion. By separating the data plane (symmetric encryption) from the control plane (public-key keying + smart contracts), the system delivers practical, privacy-preserving genomic data sharing suitable for IRB-approved cohort and clinical workflows.

bioinformatics↗

bioRxiv2025

Computer Vision Methods for Spatial Transcriptomics: A Survey

Spatial transcriptomics (ST) enables the simultaneous measurement of gene expression and spatial localization within tissue sections, providing unprecedented opportunities to dissect tissue architecture and functional organization. As a relatively new omics technology, bioinformatics has driven much of the innovation in ST. However, within these frameworks, "spatial" information is often reduced to locations and relationships between molecular profiles, without fully leveraging the wealth of submicron morphological detail and histological knowledge available. Advances in computer vision-based artificial intelligence (AI) are opening exciting new avenues beyond conventional bioinformatics approaches by modeling complex histological patterns and linking morphology to molecular states. More excitingly, they bring fresh perspectives to potentially address key limitations of ST, including its high cost, limited clinical applicability, and reliance on twodimensional (2D) analysis of inherently three-dimensional (3D) tissues. For instance, models that predict ST directly from histology images enable "virtual sequencing," drastically reducing costs while integrating morphological insights from pathology with molecular biomarkers, thus accelerating clinical translation. Moreover, computer vision techniques can reconstruct pixel-aligned 3D tissue models, overcoming the technical barriers of 2D acquisition and advancing 3D spatial omics analytics. In this paper, we present the first systematic survey of computer vision AI models for ST analytics, categorizing approaches across architectures, learning paradigms, tasks, and datasets, and tracing their technological evolution. We highlight key challenges and future directions, offering a panoramic perspective on vision-driven ST and its potential to transform both basic research and clinical practice. The curated collection of vision-driven spatial transcriptomics papers is available at https://github.com/hrlblab/computer_vision_spatial_omics.

bioinformatics↗

bioRxiv2025

GCP-VQVAE: A Geometry-Complete Language for Protein 3D Structure

AO_SCPLOWBSTRACTC_SCPLOWConverting protein tertiary structure into discrete tokens via vector-quantized variational autoencoders (VQ-VAEs) creates a language of 3D geometry and provides a natural interface between sequence and structure models. While pose invariance is commonly enforced, retaining chirality and directional cues without sacrificing reconstruction accuracy remains challenging. In this paper, we introduce GCP-VQVAE, a geometry-complete tokenizer built around a strictly SE(3)-equivariant GCPNet encoder that preserves orientation and chirality of protein backbones. We vector-quantize rotation/translation-invariant readouts that retain chirality into a 4 096-token vocabulary, and a transformer decoder maps tokens back to backbone coordinates via a 6D rotation head trained with SE(3)-invariant objectives. Building on these properties, we train GCP-VQVAE on a corpus of 24 million monomer protein backbone structures gathered from the AlphaFold Protein Structure Database. On the CAMEO2024, CASP15, and CASP16 evaluation datasets, the model achieves backbone RMSDs of 0.4377 [A], 0.5293 [A], and 0.7567 [A], respectively, and achieves 100% codebook utilization on a held-out validation set, substantially outperforming prior VQ-VAE-based tokenizers and achieving state-of-the-art performance. Beyond these benchmarks, on a zero-shot set of 1 938 completely new experimental structures, GCP-VQVAE attains a backbone RMSD of 0.8193 [A] and a TM-score of 0.9673, demonstrating robust generalization to unseen proteins. Lastly, we show that the Large and Lite variants of GCP-VQVAE are substantially faster than the previous SOTA (AIDO), reaching up to [~] 408 x and [~] 530 x lower end-to-end latency, while remaining robust to structural noise. We make the GCP-VQVAE source code, zero-shot dataset, and its pretrained weights fully open for the research community: https://github.com/mahdip72/vq_encoder_decoder

bioinformatics↗

bioRxiv2025

Zero-Shot Protein-Ligand Binding Site Prediction from Protein Sequence and SMILES

Accurate identification of protein-ligand binding sites is critical for mechanistic biology and drug discovery, yet performance varies widely across ligand families and data regimes. We present a systematic prediction and evaluation framework that stratifies ligands into three settings, overrepresented (many examples), under-represented (tens of examples; few-shot), and zero-shot (unseen at training). We developed a novel three-stage, sequence-based modeling suite that progressively adds ligand conditioning and zero-shot capability, and used an evaluation frame-work to assess the suite. Stage 1 trains per-ligand predictors using a pretrained protein language model (PLM). Stage 2 introduces ligand-aware conditioning via an embedding table, enabling a single multi-ligand model. Stage 3 replaces the table with a pretrained chemical language model (CLM) operating on SMILES, enabling zero-shot generalization. We show Stage 2 improves Macro F1 on the overrepresented test set from 0.4769 (Stage 1) to 0.5832 and outperforms sequence- and structure-based baselines. Stage 3 attains zero-shot performance (F1 = 0.3109) on 5 612 previously unseen ligands while remaining competitive on represented ligands. Ablations across five PLM scales and multiple CLMs reveal larger PLM backbones consistently increase Macro F1 across all regimes, whereas scaling the CLM yields modest or inconsistent gains, which need further investigation. Our results demonstrate that zero-shot residue-level prediction from sequence and SMILES is feasible and identifies the PLM scale as the dominant lever for further advances. The code is fully open source at GitHub.

bioinformatics↗

bioRxiv2025

CryoFSL: an annotation-efficient, few-shot learning framework for robust protein particle picking in cryo-EM micrographs

AO_SCPLOWBSTRACTC_SCPLOWAccurate identification of protein particles in cryo-electron microscopy (cryo-EM) micrographs is crucial for high-resolution structure determination, but remains challenging due to the heavy reliance on extensive annotated datasets and the difficulty of ensuring robustness under low signal-to-noise ratio (SNR) conditions. Current approaches require large annotations and exhibit poor generalization to new protein targets. We present CryoFSL (Cryo-EM Few Shot-Learning), a novel few-shot learning framework built upon Segment Anything Model 2 (SAM2) with lightweight adapters, enabling robust particle picking using as few as five labeled micrographs, significantly reducing annotation burden. The frameworks hierarchical adapter design supports dynamic feature modulation for low-SNR and heterogeneous conditions, resolving the trade-off between annotation burden and performance. CryoFSL surpasses both traditional template-based methods and state-of-the-art deep learning models across diverse proteins in the few-short learning setting, achieving superior recall, precision and 3D reconstruction resolution with minimal supervision. It maintains stability across heterogeneous micrographs and consistently detects high-quality particles with fewer false positives. Notably, CryoFSL achieves competitive density map reconstruction resolution with just a fraction of the particles picked by other methods, redefining efficiency and quality in cryo-EM analysis. This work paves the way for scalable, generalizable, and annotation-efficient particle picking pipelines. The code is available at GitHub.

bioinformatics↗

bioRxiv2025

Sphingosine kinase 2 suppresses neutrophil responses to promote viral persistence while attenuating immune pathology

Chronic virus infections often suppress immune cell functions which helps in restricting immune pathology but leads to viral persistence. However, the underlying mechanisms are incompletely understood. We recently found that sphingosine kinase 2 (SphK2)-deficient (Sphk2-/-) mice succumbed to lymphocytic choriomeningitis virus (LCMV) infection due to immune pathology. In addition to heightened T cell immunity, a notable increase of neutrophils was observed in LCMV-infected Sphk2-/- mice. Depletion of neutrophils increased the viability of virus-infected Sphk2-/- mice, supporting a role of SphK2-deficient neutrophils in viral immune pathogenesis. Further, SphK2-deficient neutrophils expressed lower levels of the immune suppressive marker CD244 during infection. Importantly, adoptively transferred SphK2-deficient neutrophils demonstrated intrinsic regulation of CD244 and improved virus-specific T cell responses, resulting in diminished viral burden. Transcriptomic analysis revealed increased expression of pro-inflammatory and antiviral genes in SphK2-deficient neutrophils. These results indicate that SphK2 promotes suppressive neutrophil responses and regulates neutrophil-associated immune pathology during a persistent infection. Our findings may help design new immune therapeutics to control chronic viral diseases. SignificanceNeutrophils are the sentinels of the innate immune system; they can reshape innate and adaptive immune responses. During chronic illnesses, such as persistent viral infections, neutrophils can suppress the host immune response and help in disease progression. Here, we demonstrate regulation of neutrophil expansion and functions by sphingosine kinase 2 (SphK2) during LCMV infection. SphK2-deficient neutrophils express a reduced level of inhibitory receptor CD244, exert immune stimulatory effects on T cells, and promote virus clearance. Further, transcriptomic analysis reveals that SphK2 deficiency leads to the development of proinflammatory neutrophils. Our study identifies SphK2, a host factor, as being critical for neutrophil suppression that regulates dysfunctional T cell response and virus persistence.

immunology↗

bioRxiv2025

The anti-phage mechanism of a widespread trypsin-MBL module

Protease mediate activation of immune effectors is a conserved mechanism across life. This study identified a widespread and co-evolving trypsin-MBL module as a core effector in diverse bacterial immune systems (e.g., Hachiman, Argonaute, AVAST), protecting host via abortive infection. Focusing on Hachiman-trypsin-MBL, we demonstrated that, the protease activity of trypsin*HamAB is inhibited in the presence of ATP and MBL is an autoinhibited DNase with two inserted loops blocking the catalytic pocket. Upon infection, the HamB helicase senses invading nucleic acids and catalyzes the hydrolysis of ATP, activating the associated trypsin-like protease that cleaves specifically at two non-canonical KSS motif of MBL and releases its DNase activity. Cryo-EM structures of activated trypsin*HamAB bound with DNA reveal DNA binding and ATP hydrolysis lead to the release of the trypsin-like domain, enabling activation. Our work provides a comprehensive understanding on the mechanism of widely distributed trypsin-MBL module, wherein proteolytic activation of a toxic nuclease enables robust bacterial immunity while preventing self-toxicity through multi-layered control.

biochemistry↗

bioRxiv2025

ProtLoc-GRPO: Cell line-specific subcellular localization prediction using a graph-based model and reinforcement learning

Subcellular localization prediction is crucial for understanding protein functions and cellular processes. Subcellular localization is dependent on tissue and cell lines derived from different cell types. Predicting cell line-specific subcellular localization using the information of protein-protein interactions (PPIs) offers deeper insights into dynamic cellular organization and molecular mechanisms. However, many existing PPI networks contain systematic errors that limit prediction accuracy. In this study, we propose a reinforcement learning approach, ProtLoc-GRPO, to enhance subcellular localization prediction by optimizing the structure of the underlying PPI network. ProtLoc-GRPO learns to rank and retain the most informative PPI edges to maximize the macro-F1 score for cell line-specific subcellular localization. Our approach yields a 7% improvement in macro-F1 score over the baseline. We further evaluate its robustness across various edge pruning rates and benchmark it against conventional pruning strategies. Results show that our proposed method consistently outperforms existing approaches. To our knowledge, this work represents the first study to predict cell line-specific protein subcellular localization and the first application of the Group Relative Policy Optimization (GRPO) framework to a graph-based model for bioinformatics tasks.

bioinformatics↗

bioRxiv2025

MiroSCOPE: An AI-driven digital pathology platform for annotating functional tissue units

Cancer tissue analysis in digital pathology is typically conducted across different spatial scales, ranging from high-resolution cell-level modeling to lower-resolution tile-based assessments. However, these perspectives often overlook the structural organization of functional tissue units (FTUs), the small, repeating structures which are crucial to tissue function and key factors during pathological assessment. The incorporation of FTU information is hindered by the need for detailed manual annotations, which are costly and time-consuming to obtain. While artificial intelligence (AI)-based solutions hold great promise to accelerate this process, there is currently no comprehensive workflow for building the large, annotated cohorts required. To remove these roadblocks and advance the development of more interpretable approaches, we developed MiroSCOPE, an end-to-end AI-assisted platform for annotating FTUs at scale, built on QuPath. MiroSCOPE integrates a fine-tunable multiclass segmentation model and curation-specific usability features to enable a human-in-the-loop system that accelerates AI annotation by a pathologist. The system is used to efficiently annotate over 71,900 FTUs on 184 prostate cancer hematoxylin and eosin (H&E)-stained tissue samples and demonstrates ready translation to breast cancer. Furthermore, we publicly release a dataset named Miro-120, consisting of 120 prostate cancer H&E with 30,568 annotations, which can be used by the community as a high-quality resource for FTU-level machine learning aims. In summary, MiroSCOPE provides an adaptable AI-driven platform for annotating functional tissue units, facilitating the use of structural information in digital pathology analyses.

bioinformatics↗

bioRxiv2025

Global analysis of membrane protein S-acylation in the model plant Arabidopsis thaliana

Protein S-acylation is the addition of fatty acids to the cysteine residues in a protein, catalyzed by protein S-acyltransferases (PATs). Despite extensive research on protein S-acylation in animals, our understanding of this process in plants remains limited. In this study, we sought to characterize the S-acylproteome of membrane proteins in Arabidopsis and identify potential substrates for two important plant immunity-related PATs (PAT5 and PAT9). To achieve this, S-acylated membrane proteins were first enriched via our optimized acyl-biotinyl exchange strategies at both the protein-level and peptide-level. The enriched samples were then analyzed by label-free quantitative liquid chromatography-mass spectrometry. The results from the two enrichment methods demonstrated that they were complementary in identifying S-acylated proteins and S-acylation sites. Using these methods, over 2500 S-acylation sites in more than 2000 putative S-acylated proteins were identified. Proteins involved in vesicle trafficking, plant phosphorylation, immune responses, and signal transduction pathways were significantly enriched. Additionally, certain amino acid patterns surrounding the S-acylation sites were identified. Comparisons of the S-acylproteomes between the wild type and the PAT5 and PAT9 mutants revealed over 100 potential substrates for both S-acyltransferases. The high quality of our data was supported by the significant overlap with the previously reported data and successful experimental verification of selected candidate proteins. Overall, our study revealed a well-represented S-acylproteome for Arabidopsis (especially its membrane proteins) and identified potential substrates for PAT5 and PAT9. These findings will facilitate the functional characterization of S-acylated proteins in plants.

plant biology↗

bioRxiv2025

Dynamics and Regulation of mRNA Cap Recognition by Human eIF4F

Efficient eukaryotic messenger RNA translation requires dynamic collaboration between the three subunits of initiation factor 4F (eIF4F, eIF4E*G*A), which recognises and activates mRNA at its 5' cap structure for ribosome recruitment. Despite its high biological and pharmacological importance, the dynamics of full human eIF4F-mRNA engagement remain largely uncharacterised, hindering mechanistic understanding of translation initiation and its regulation. Here we observed human eIF4F activity with single-molecule fluorescence assays that directly visualise mRNA cap recognition by its eIF4E subunit. Unexpectedly, we find that inherently transient eIF4E-cap binding is repressed by full-length human eIF4G, predominantly through its C-terminus, representing an unanticipated role for eIF4G as a central rate-limiting factor in the eIF4F complex. This repression is relieved by nucleotide-bound eIF4A in the eIF4F heterotrimer, placing eIF4A as a crucial determinant of efficient cap recognition for translation. Molecular dynamics simulations reveal that electrostatic modulation of eIF4E-mRNA interaction allows eIF4G to control the cap-recognition frequency. Our findings also indicate that intrinsic eIF4F- mRNA dynamics are insufficient to support cap-tethered ribosomal scanning to locate translation start sites. They illuminate fundamental design-principle differences for the overall mechanism and division of labour among eIF4F subunits during mRNA recognition in humans and yeast.

biophysics↗

bioRxiv2025

Aptamer-Enabled Discovery and Clinical Analysis of Exosomal Surface Biomarkers in Hepatocellular Carcinoma

Exosomal membrane proteins hold promise as cancer biomarkers but face detection challenges due to low abundance and conformational sensitivity. We developed AptEx-ID, a microbead-displayed aptamer platform enabling high- throughput screening of hepatocellular carcinoma (HCC)-specific exosomal surface proteins. Aptamers, selected via competitive binding to tumor-derived versus healthy exosomes, demonstrated high affinity and specificity. Clinical validation using 50 HCC and 30 control plasma samples identified a 7-aptamer panel via machine learning, achieving 93.75% accuracy and 0.95 ROC-AUC. These aptamers further revealed HCC-associated biomarkers (e.g., IMPDH1, ASCC3, GTSE1), validated by siRNA knockdown and biophysical assays. Post-treatment monitoring showed aptamer signal reduction in responsive patients, highlighting their prognostic utility. This platform integrates biomarker discovery, diagnostics, and therapeutic monitoring, offering a robust approach for HCC management through exosomal surface proteome profiling.

molecular biology↗

bioRxiv2025

Deep-learning-enabled multi-omics analyses for prediction of future metastasis in cancer

Metastasis remains the leading cause of cancer-related mortality, yet predicting future metastasis is a major clinical challenge due to the lack of validated biomarkers and effective assessment methods. Here, we present EmitGCL, a deep-learning framework that accurately predicts future metastasis and its corresponding biomarkers. Based on a comprehensive benchmarking comparison, EmitGCL outperformed other computational tools across six cancer types from seven cohorts of patients with superior sensitivity and specificity. It captured occult metastatic cells in a patient with a lymph node-negative breast cancer, who was declared to have no evidence of disease by conventional imaging methods but was later confirmed to have a metastatic disease. Notably, EmitGCL identified HSP90AA1 and HSP90AB1 as predictable biomarkers for future breast cancer metastasis, which was validated across five independent cohorts of patients (n=420). Furthermore, we demonstrated YY1 transcription factor as a key driver of breast cancer metastasis which was validated through in-silico and CRISPR-based migration assays, suggesting that YY1 is a potential therapeutic target for deterring metastasis.

bioinformatics↗

bioRxiv2025

Disentangling complex language contact and admixture in the broad Gansu-Qinghai region

Language evolution in the Gansu-Qinghai (GQ) region provides a key perspective for understanding cultural development along the Silk Road. Previous genetic and archaeological studies revealed complex multiethnic interactions shaped by migration and sociocultural exchange. However, the lack of structured linguistic data and computational tools for studying language contact has limited rigorous analysis of sociocultural evolution in this area. Here, we presented a new hybrid dataset of phonological and morpho-syntactic features from languages sampled across the GQ region. We introduced a computational framework to assess language contact and admixture, allowing us to quantify interaction among GQ languages and trace their origins. Our results showed that GQ languages have distinct contact patterns in their phonological and morphosyntactic systems, with some languages exhibiting clear evidence of mixture. Using a new statistical method, Trait Sharing Test (TS-TEST), based on tree topology, we identified significant influences from Sinitic, Tibetan, Mongolic, and Turkic languages in shaping GQ linguistic diversity. These findings highlight the GQ region as a linguistic convergence zone on the Silk Road and provide a foundation for quantitative research on language contact and admixture. Our work strengthens the linguistic perspective on cultural evolution in the GQ region and supports future interdisciplinary studies integrating languages, genes, and material cultures.

evolutionary biology↗

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↗