Biology subjects

Zhong, B.

Publications and source records attributed to Zhong, B..

At least 19 recordsLinked to original sources

bioRxiv2026

VenusGenome: Genome-Aware Masked Autoencoding for Microbial Genome Representation Learning

Genomic-context models have shown that contextualizing protein embeddings can improve predictions of microbial function. We observe that a gene's protein embedding is partially predictable from the embeddings of neighboring genes in the genome, motivating high-ratio masking that requires reconstruction from sparser genomic context. Genomic modeling must also distinguish biological relationships from assembly conventions, since the ordering of separate contigs does not imply adjacency and strand labels depend on the reference orientation. We therefore introduce VenusGenome, a gene-resolution masked autoencoder that combines visible-only encoding with genome-aware attention incorporating contig boundaries, relative strand orientations, and gene positions. We pretrain this model on a curated corpus of over 1.7 million microbial genome assemblies from isolate and metagenomic sources, using frozen protein embeddings as gene-level inputs. Across genome-level phenotype prediction and gene-level tasks, the resulting frozen representations achieve the highest mean performance among the evaluated models on nine of eleven primary metrics. An exploratory gene-input perturbation analysis further links representation sensitivity to experimentally measured fitness defects. Ablations support the contribution of genome-aware attention and show the benefits of masked autoencoding over BERT-style masked modeling, with further gains from higher masking ratios.

bioinformatics↗

bioRxiv2026

Bayesian inference of global land plant diversification dynamics across the Phanerozoic

Underlying diversification patterns and potential drivers of land plants remain puzzling, particularly at deep-time and global scales. Here, we reconstruct the diversification dynamics of land plants using an updated time-calibrated phylogeny comprising approximately 30,000 extant species, and nearly 40,000 fossil occurrences within the Bayesian birth-death frameworks. Our analyses indicate that angiosperms exhibit the highest diversification rates, whereas ferns underwent a marked diversification increase during the Cretaceous, coincident with the rapid radiation of angiosperms. Among the major Phanerozoic mass extinctions, the Cretaceous-Paleogene event has the most significant impact on land plant diversity, with estimated genus-level losses of 13%, 14%, and 4% in bryophytes, pteridophytes, and gymnosperms. We identify a pattern consistent with competitive displacement scenario, in which pteridophytes origination rates declined as gymnosperm diversity increased, accompanying the transition from pteridophyte- to gymnosperm-dominated floras. We find that plant-insect interactions have originated as early as the emergence of pteridophytes, potentially promoting the diversification of pteridophytes and gymnosperms. During the Cretaceous, these associations shifted toward angiosperms, as indicated by a positive relationship between insect diversity and angiosperm origination rates. Finally, continental fragmentation, fluctuations in aridity, atmospheric carbon and sedimentary dynamics, showed stronger associations with the successive rise and decline of major land plant groups.

evolutionary biology↗

bioRxiv2026

Deep learning-guided design of hydrolases for crystalline PET depolymerization

Poly(ethylene terephthalate) (PET), a ubiquitous polyester used in packaging and textiles, persists in the environment due to its high crystallinity and stability, contributing substantially to global plastic pollution. Enzymatic depolymerization by PET hydrolase (PETase) offers a chemically precise and environmentally sustainable route to convert PET into its monomeric building blocks, enabling recycling. However, practical implementation remains hindered by the rigid, crystalline architecture of PET, which restricts enzyme access and necessitates energy-intensive pretreatment to enable efficient depolymerization. In addition, most PETases achieve only partial depolymerization with low terephthalic acid (TPA) yields and accumulate inhibitory intermediates, while limited thermostability and slow surface kinetics further restrict efficiency. To overcome these barriers, we report VenusPETase, an engineered variant of KbPETase designed using PET-Flow, a state-of-the-art computational framework for PETase engineering. Compared with KbPETase, VenusPETase exhibits a 3.4-fold increase in hydrolytic activity, up to a 27-fold improvement after heat treatment of protein, and a 12 {degrees}C enhancement in thermostability. VenusPETase exhibits rapid degradation across a wide crystallinity range (8%-50%) at 50 {degrees}C, effectively spanning the entire spectrum of commercial PET products. Moreover, VenusPETase outperformed nine high-performance PETases under their respective optimal conditions and degraded untreated PET substrates across 8%-50% crystallinity, producing TPA as over 95% of the released products with minimal intermediate accumulation. X-ray crystallography and molecular dynamics simulations suggest that dynamic modulation, elevated surface electrostatic potential enhance the interaction of VenusPETase with crystalline PET, thereby lowering the hydrolytic energy barrier and improving catalytic performance. We also demonstrate that untreated, postconsumer-PET from nine different products can all be degraded by VenusPETase. The recovered monomers can be directly repolymerized into virgin-quality PET, demonstrating a closed-loop enzymatic recycling process. In a 100 L bioreactor, VenusPETase completely depolymerizes post-consumer crystalline PET (28% crystallinity) within 24 h under 50 {degrees}C. These results establish VenusPETase as a robust biocatalyst that enables efficient, closed-loop recycling of crystalline PET under mild conditions.

biochemistry↗

bioRxiv2026

An additional site mutation in MITA/STING gain-of-function mutants abolishes the autoimmune SAVI phenotypes and directs a therapeutic strategy.

STING-associated vasculopathy with onset in infancy (SAVI) is an autoimmune disease caused by gain-of-function mutations (GOFs) of MITA/STING and the most frequent GOFs for SAVI are MITAN154S and MITAV155M. However, how MITA GOFs are spontaneously activated remains incompletely understood. Here, we show that the activity of MITA hinge-region GOFs is compromised by an additional mutation at Lys150 and that the SAVI phenotypes of MITAN153S/WT mice are completely abolished in the MITAK150N/N153S (MITANS/NS) mice. Mechanistically, MITA GOFs constitutively associate with iRhom2 for the spontaneous ER-to-Golgi translocation, which is substantially inhibited by the introduction of a mutation at Lys150. Interestingly, cGAMP binds to MITANS, triggers the ER-to-Golgi translocation of MITANS as well as the MITANS-iRhom2 interaction, and induces the expression of downstream genes in MitaNS/NS cells similarly as in Mita+/+ cells. Consistently, structural studies demonstrate an inactive open conformation of apo-MITANS characterized by connector region crossover and a curved filament of cGAMP-bound MITANS characterized by parallel connector regions, similar to those observed in wild-type MITA. Furthermore, we design a SAVI-inhibitory peptide (SIP) that selectively inhibits the interaction between MITAN153S and iRhom2 and the activity of MITA GOFs and thereby abolishes the SAVI phenotypes of the MITAN153S/WT[->]WT chimeric mice. These findings reveal a previously uncharacterized mechanism for the spontaneous activation of MITA GOFs and highlight a potential therapeutic intervention for SAVI.

immunology↗

bioRxiv2025

Tumor cell-specific loss of GPX4 reprograms triacylglycerol metabolism to escape ferroptosis and impair antitumor immunity in NSCLC

Glutathione peroxidase 4 (GPX4) is a master regulator of ferroptosis, a process that has been proposed as a potential therapeutic strategy for cancer. Here we have unexpectedly found that inducible knockout of GPX4 in tumor cells significantly promotes non-small cell lung cancer (NSCLC) progression in the autochthonous KrasLSL-G12D/+Lkb1fl/fl (KL) and KrasLSL-G12D/+Tp53fl/fl (KP) mouse models, whereas inducible overexpression of GPX4 in tumor cells exerts the opposite effect. GPX4-deficient tumor cells evade ferroptosis by upregulating the expression of DGAT1/2 to promote the synthesis of triacylglycerol (TAG) and oxidized TAG (oxTAG) and the formation of lipid droplets in cells. In addition, GPX4-deficient tumor cells secrete TAG and oxTAG into the extracellular space to induce dysfunction of antitumor CD8+ T cells, thereby coordinating an immunoinhibitory tumor microenvironment (TME). Consistently, treatment with DGAT1/2 inhibitors or inducible overexpression of GPX4 in tumor cells significantly resensitizes tumor cells to ferroptosis and ignites the activation of T cells in the TME to inhibit NSCLC progression. These findings highlight a previously uncharacterized role of tumor cell-specific GPX4 in NSCLC progression and provide potential therapeutic strategies for NSCLC.

cell biology↗

bioRxiv2025

Venus-MAXWELL: Efficient Learning of Protein-Mutation Stability Landscapes using Protein Language Models

In-silico prediction of protein mutant stability, measured by the difference in Gibbs free energy change ({Delta}{Delta}G), is fundamental for protein engineering. Current sequence-to-label methods typically employ the two-stage pipeline: (i) encoding mutant sequences using neural networks (e.g., transformers), followed by (ii) the {Delta}{Delta}G regression from the latent representations. Although these methods have demonstrated promising performance, their dependence on specialized neural network encoders significantly increases the complexity. Additionally, the requirement to individually compute latent representations for each mutant site negatively impacts computational efficiency and poses the risk of overfitting. This work proposes the Venus-MO_SCPLOWAXWELLC_SCPLOW framework, which reformulates mutation {Delta}{Delta}G prediction as a sequence-to-landscape task. In Venus-MO_SCPLOWAXWELLC_SCPLOW, mutations of a protein and their corresponding {Delta}{Delta}G values are organized into a landscape matrix, allowing our framework to learn the {Delta}{Delta}G landscape of a protein with a single forward and backward pass during training. Besides, to facilitate future works, we also curated a large-scale {Delta}{Delta}G dataset with strict controls on data leakage and redundancy to ensure robust evaluation. Venus-MO_SCPLOWAXWELLC_SCPLOW is compatible with multiple protein language models and enables these models for accurate and efficient {Delta}{Delta}G prediction. For example, when integrated with the ESM-IF, Venus-MO_SCPLOWAXWELLC_SCPLOW achieves higher accuracy than ThermoMPNN with 10x faster in inference speed (despite having 50x more parameters than ThermoMPNN). The training codes, model weights, and datasets are publicly available at https://github.com/ai4protein/Venus-MAXWELL.

bioinformatics↗

bioRxiv2025

LOSS OF ROR2 TYROSINE KINASE RECEPTOR IS ASSOCIATED WITH ENDOTHELIAL DYSFUNCTION IN PAH VIA INAPPROPRIATE INTEGRIN BETA 1 ACTIVATION

RationaleEndothelial dysfunction is a key feature of pulmonary arterial hypertension (PAH). We previously identified Wnt7a, a ligand of the Wnt planar cell polarity (PCP) pathway, as essential for pulmonary angiogenesis, with its loss linked to PAH. Given the importance of Wnt/PCP to lung endothelial function and angiogenesis, our goal is to elucidate how Wnt/PCP regulates angiogenic responses in pulmonary microvascular endothelial cells (PMVECs). ROR2, a tyrosine kinase receptor specific to Wnt/PCP, is crucial for cardiovascular development, but its role in PAH is unclear. We hypothesized that ROR2 supports endothelial homeostasis, and its loss would impair angiogenesis, contributing to PAH. MethodsEndothelial-specific ROR2 knockout (ROR2 ECKO) and wild-type (WT) mice were studied under normoxia and chronic hypoxia using echocardiography, hemodynamics, and lung morphometry. PMVECs from healthy and PAH lungs were transfected with ROR2 siRNA/constructs for functional and molecular studies. Focal adhesion (FA) activation and force generation were assessed via FRET-based methods. Bulk and single-cell transcriptomic analyses were performed on siROR2 PMVECs and ROR2 ECKO lungs. ResultsROR2 ECKO mice exhibited worsened pulmonary hypertension, right ventricular remodeling, microvascular loss, and muscularization in hypoxia. Single-cell RNA sequencing of lung endothelial cells showed dysregulation of pathways involved in barrier formation and angiogenesis. Evans blue dye extravasation confirmed reduced endothelial barrier integrity in ROR2 ECKO mice. ROR2-deficient PAH PMVECs displayed increased adhesion, permeability, and FA numbers, with reduced VE-cadherin at cell junctions. Confocal imaging revealed ROR2 localization in FAs, interacting with integrin {beta}1 (ITGB1). FRET analysis showed that ITGB1 remained in an active, adhesion-promoting state in ROR2-deficient cells. Restoring ROR2 in PAH PMVECs normalized adhesion, barrier function, and FA abundance. Transcriptomic analysis identified Rab12 as a key mediator of ROR2-ITGB1 crosstalk, with Rab12 knockdown mimicking ROR2 deficiency in PMVECs. ConclusionsROR2 regulates pulmonary angiogenesis by maintaining endothelial barrier integrity and facilitating integrin recycling. Restoring ROR2 signaling could be a potential therapeutic approach for PAH.

cell biology↗

bioRxiv2025

Unveiling Protein-DNA Interdependency: Harnessing Unified Multimodal Sequence Modeling, Understanding and Generation

The interdependent orchestration of proteins, coding DNA sequences (CDS), and adjacent regulatory DNA elements underpins biological functionality, where evolutionary constraints shape codon usage, transcriptional regulation, and structural stability. We present ProDMM, a multimodal framework that deciphers these systemic relationships through unified sequence modeling. Integrating a context-aware encoder with conditional generation, our architecture captures codependent patterns between protein motifs and cis-regulatory syntax. ProDMMs encoder demonstrates superior predictive capabilities in forecasting pathway-level phenotypes and protein expression profiles, outperforming conventional single-modality approaches. Its decoder generates synthetically optimized regulatory elements by resolving spatial dependencies between coding sequences and adjacent non-coding regions, enabling transcriptional fine-tuning unavailable to sequence-agnostic methods. By resolving how localized CDS-NCDS-protein interdependencies propagate to system-level phenotypes, ProDMM establishes a paradigm for holistic biological sequence engineering. This approach enables coordinated optimization of translation efficiency, enzymatic activity, and pathway flux, offering transformative potential for designing tailored biosynthesis systems and advancing sustainable bioproduction strategies.

bioinformatics↗

bioRxiv2025

SCOIGET: Predicting Spatial Tumor Evolution Pattern by Inferring Spatial Copy Number Variation Distributions

Constructing a comprehensive spatiotemporal map of tumor heterogeneity is essential for understanding tumor evolution, with copy number variation (CNV) as a significant feature. Existing studies often rely on tools originally developed for single-cell data, which fail to utilize spatial information. Our study aims to develop a model that fully leverages spatial omics data to elucidate spatio-temporal changes in tumor evolution. Here, we introduce SCOIGET (Spatial COpy number Inference by Graph on Evolution of Tumor), a novel framework using graph neural networks with graph attention layers to learn spatial neighborhood features of gene expression and infer copy number variations. This approach integrates spatial multi-omics features to create a comprehensive spatial map of tumor heterogeneity. Notably, SCOIGET achieves a substantial reduction in error metrics (e.g., mean squared error, cosine similarity, and distance measures) and produces superior clustering performance as indicated by higher Silhouette Scores compared to existing methods. Our model significantly enhances the efficiency and accuracy of tumor evolution depiction, capturing detailed spatial and temporal changes within the tumor microenvironment. It is versatile and applicable to various downstream tasks, demonstrating strong generalizability across different spatial omics technologies and cancer types. This robust performance improves research efficiency and provides valuable insights into tumor progression. In conclusion, SCOIGET offers an innovative solution by integrating multiple features and advanced algorithms, providing a detailed and accurate representation of tumor heterogeneity and evolution, aiding in the development of personalized cancer treatment strategies. Key PointsO_ST_ABS1. Innovative Framework for Spatially Contextualized CNV InferenceC_ST_ABSIntroduces SCOIGET, a graph-based model that integrates spatial omics data with graph neural networks and attention mechanisms to accurately infer spatial copy number features. 2. Detailed Tumor Heterogeneity MappingConstructs representations of tumor evolution, capturing heterogeneity and clonal dynamics at both cellular and subcellular resolutions. 3. Versatile and Robust PerformanceOutperforms existing methods across multiple datasets, demonstrating reliability and adaptability to diverse spatial omics platforms and cancer types. 4. Broad Downstream ApplicationsEnables tasks such as tumor subclone identification, clustering, and evolutionary trajectory analysis, facilitating precision oncology strategies.

bioinformatics↗

bioRxiv2025

D-LMBmapX: Generalized Deep-Learning Pipeline for Automated Whole-brain Neural Circuitry Profiling Across Any Developmental Stages

Understanding whole-brain circuitry development is essential for uncovering the origins of disorders linked to abnormal neural wiring, yet progress has been limited by the lack of tools for accurate, comprehensive analysis. Brain development is highly dynamic and region-specific, with pronounced inter-individual variability during early stages, and no existing method achieves whole-brain profiling at arbitrary stages. The rapidly changing morphology of developing axons further complicates segmentation. Here, we present D-LMBmapX, a deep learning pipeline for automated whole-brain circuitry profiling across postnatal development. D-LMBmapX constructs sample-inferred atlases from flanking anchor stages, enabling accurate registration at any time point. By developing a foundation model for generalised axon and soma segmentation, it supports quantitative mesoscale developmental connectomics, demonstrated through spatial-temporal profiling of catecholaminergic projections. D-LMBmapX also enables robust cross-modality and cross-dimensional registration, including precise alignment of single 2D slices to 3D references, and offers generalizable strategies for 5D analysis of diverse biomedical spatial datasets.

neuroscience↗

bioRxiv2025

Discovery of Expression-Governing Residues in Proteins

Understanding how amino acids influence protein expression is crucial for advancements in biotechnology and synthetic biology. In this study, we introduce Venus-TIGER, a deep learning model designed to accurately identify amino acids critical for expression. By constructing a two-dimensional matrix that links model representations to experimental fitness, Venus-TIGER achieves improved predictive accuracy and enhanced extrapolation capability. We validated our approach on both public deep mutational scanning datasets and low-throughput experimental datasets, demonstrating notable performance compared to traditional methods. Venus-TIGER exhibits robust trans-ferability in zero-shot predicting scenarios and enhanced predictive performance in few-shot learning, even with limited experimental data. This capability is particularly valuable for protein design aimed at enhancing expression, where generating large datasets can be costly and time-consuming. Additionally, we conducted a statistical analysis to identify expression-associated features, such as sequence and structural preferences, distinguishing between those linked to high and low expression. Our investigation also revealed a correlation among stability, activity and expression, providing insight into their interconnected roles and underlying mechanisms.

bioengineering↗

bioRxiv2024

Harnessing Protein Language Model for Structure-Based Discovery of Highly Efficient and Robust PET Hydrolases

Plastic waste, particularly polyethylene terephthalate (PET), presents significant environmental challenges, prompting extensive research into enzymatic biodegradation. Existing PET hydrolases are limited to a narrow sequence space and demonstrate insufficient performance for biodegradation. This study introduces a novel discovery pipeline that combines protein language models (PLMs) with a structural representation tree to identify enzymes based on structural similarity. Using the crystal structure of IsPETase as a template, we employed PLMs and a representation tree to efficiently search and cluster target proteins. Screening of candidate proteins was further refined using PLM-based assessments of solubility and thermal stability. Biochemical experiments showed that 14 of 34 candidates exhibited PET degradation activity across a temperature range of 30-60 {degrees}C. Notably, we identified a PET hydrolase -KbPETase, which possesses a melting temperature 32 {degrees}C higher than that of IsPETase and exhibits the highest PET degradation activity within 30-50 {degrees}C compared to other wild-type PETases. KbPETase also shows higher catalytic efficiency than FastPETase. X-ray crystallography and molecular dynamics simulations further revealed that KbPETase has a conserved catalytic domain and enhanced intramolecular interactions. This work develops a novel deep learning approach to discover natural PETases with enhanced functions.

molecular biology↗

bioRxiv2024

How to select the best zero-shot model for the viral proteins?

Predicting the fitness of viral proteins holds notable implications for understanding viral evolution, advancing fundamental biological research, and informing drug discovery. However, the considerable variability and evolution of viral proteins make predicting mutant fitness a major challenge. This study introduces the ProPEC, a Perplexity-based Ensemble Model, aimed at improving the performance of zero-shot predictions for protein fitness across diverse viral datasets. We selected five representative pretrained language models (PLMs) as base models. ProPEC, which integrates perplexity-weighted scores from these PLMs with GEMME, demonstrates superior performance compared to individual models. Through parameter sensitivity analysis, we highlight the robustness of perplexity-based model selection in ProPEC. Additionally, a case study on T7 RNA polymerase activity dataset underscores ProPECs predictive capabilities. These findings suggest that ProPEC offers an effective approach for advancing viral protein fitness prediction, providing valuable insights for virology research and therapeutic development. TOC Graphic O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=111 SRC="FIGDIR/small/616860v1_ufig1.gif" ALT="Figure 1"> View larger version (28K): org.highwire.dtl.DTLVardef@30be8forg.highwire.dtl.DTLVardef@2ebf35org.highwire.dtl.DTLVardef@10b517eorg.highwire.dtl.DTLVardef@134681_HPS_FORMAT_FIGEXP M_FIG C_FIG

bioinformatics↗

bioRxiv2024

Precise Generation of Conformational Ensembles for Intrinsically Disordered Proteins Using Fine-tuned Diffusion Models

Intrinsically disordered proteins (IDPs) play pivotal roles in various biological functions and are closely linked to many human diseases including cancer, diabetes and Alzheimer disease. Structural investigations of IDPs typically involve a combination of molecular dynamics (MD) simulations and experimental data to correct for intrinsic biases in simulation methods. However, these simulations are hindered by their high computational cost and a scarcity of experimental data, severely limiting their applicability. Despite the recent advancements in structure prediction for structured proteins, understanding the conformational properties of IDPs remains challenging partly due to the poor conservation of disordered protein sequences and limited experimental characterization. Here, we introduce IDPFold, a method capable of generating conformational ensembles for IDPs directly from their sequences using fine-tuned diffusion models. IDPFold bypasses the need for Multiple Sequence Alignments (MSA) or experimental data, achieving accurate predictions of ensemble properties across numerous IDPs. By sampling conformations at the backbone level, IDPFold provides more detailed structural features and more precise property estimation compared to other state-of-the-art methods. IDPFold is ready to be used in the elucidate the sequence-disorder-function paradigm of IDPs.

bioinformatics↗

bioRxiv2024

MedImg: a database for public medical image integration

The advancements of deep learning algorithms in medical image analysis has garnered tremendous attention in recent years. Several studies have reported that the models have achieved and even surpassed human performance, whereas the translation of these models into clinical practice is still accompanied by various challenges. A major challenge is the large-scale and well characterized dataset to validate the generalization of models. Therefore, we collected diverse medical image datasets from multiple public sources containing 103 datasets, 1,622,956 images. These images are derived from 14 modalities like XR, CT, MRI, OCT, ultrasound, and endoscopy, and from 9 organs such as lung, brain, eye, and heart. Subsequently, we constructed an online database, MedImg, which incorporates and hierarchically organizes medical images to facilitate data access. MedImg serves as an intuitive and open-access platform for contributing to deep learning-based medical image analysis, accessible at https://www.cuilab.cn/medimg/.

bioinformatics↗

bioRxiv2024

DeProt: A protein language model with quantizied structure and disentangled attention

Protein language models (PLMs) have shown remarkable capabilities in various protein function prediction tasks. However, while protein function is intricately tied to structure, most existing PLMs do not incorporate protein structure information. To address this issue, we introduce ProSST, a Transformer-based protein language model that seamlessly integrates both protein sequences and structures. ProSST incorporates a structure quantization module and a Transformer architecture with disentangled attention. The structure quantization module translates a 3D protein structure into a sequence of discrete tokens by first serializing the protein structure into residue-level local structures and then embeds them into dense vector space. These vectors are then quantized into discrete structure tokens by a pre-trained clustering model. These tokens serve as an effective protein structure representation. Furthermore, ProSST explicitly learns the relationship between protein residue token sequences and structure token sequences through the sequence-structure disentangled attention. We pre-train ProSST on millions of protein structures using a masked language model objective, enabling it to learn comprehensive contextual representations of proteins. To evaluate the proposed ProSST, we conduct extensive experiments on the zero-shot mutation effect prediction and several supervised downstream tasks, where ProSST achieves the state-of-the-art performance among all baselines. Our code and pretrained models are publicly available 2.

bioinformatics↗

bioRxiv2024

Scaffold-Lab: Critical Evaluation and Ranking of Protein Backbone Generation Methods in A Unified Framework

De novo protein design has undergone a rapid development in recent years, especially for backbone generation, which stands out as more challenging yet valuable, offering the ability to design novel protein folds with fewer constraints. However, a comprehensive delineation of its potential for practical application in protein engineering remains lacking, as does a standardized evaluation framework to accurately assess the diverse methodologies within this field. Here, we proposed Scaffold-Lab benchmark focusing on evaluating unconditional generation across metrics like designability, novelty, diversity, efficiency and structural properties. We also extrapolated our benchmark to include the motif-scaffolding problem, demonstrating the utility of these conditional generation models. Our findings reveal that FrameFlow and RFdiffusion in unconditional generation along with Rfdiffusion and GPDL in conditional generation showcased the most outstanding performances. Furthermore, we described a systematic study to investigate conditional generation and applied it to the motif-scaffolding task, offering a novel perspective for the analysis and development of conditional protein design methods. All data and scripts will be available at https://github.com/Immortals-33/Scaffold-Lab.

bioinformatics↗

bioRxiv2023

Conditional Protein Denoising Diffusion Generates Programmable Endonucleases

Deep learning-based methods for generating functional proteins address the growing need for novel biocatalysts, allowing for precise tailoring of functionalities to meet specific requirements. This emergence leads to the creation of highly efficient and specialized proteins with wide-ranging applications in scientific, technological, and biomedical domains. This study establishes a pipeline for protein sequence generation with a conditional protein diffusion model, namely CPDiffusion, to deliver diverse sequences of proteins with enhanced functions. CPDiffusion accommodates protein-specific conditions, such as secondary structure and highly conserved amino acids (AAs). Without relying on extensive training data, CPDiffusion effectively captures highly conserved residues and sequence features for a specific protein family. We applied CPDiffusion to generate artificial sequences of Argonaute (Ago) proteins based on the backbone structures of wild-type (WT) Kurthia massiliensis Ago (KmAgo) and Pyrococcus furiosus Ago (PfAgo), which are complex multi-domain programmable endonucleases. The generated sequences deviate by up to nearly 400 AAs from their WT templates. Experimental tests demonstrated that the majority of generated proteins show unambiguous activity in DNA cleavage for both KmAgo and PfAgo, with many of them exhibiting superior activity as compared to the WT. These findings underscore CPDiffusions remarkable success rate to generate novel sequences for proteins of complex structures and functions in a single step with enhanced activity. This approach facilitates the design of enzymes with multi-domain molecular structures and intricate functions through in silico generation and screening, all accomplished without any supervision from labeled data.

bioengineering↗