Bacterial outer membrane vesicle nanorobot.

Autonomous nanorobots represent an advanced tool for precision therapy to improve therapeutic efficacy. However, current nanorobotic designs primarily rely on inorganic materials with compromised biocompatibility and limited biological functions. Here, we introduce enzyme-powered bacterial outer membrane vesicle (OMV) nanorobots. The immobilized urease on the OMV membrane catalyzes the decomposition of bioavailable urea, generating effective propulsion for nanorobots. This OMV nanorobot preserves the unique features of OMVs, including intrinsic biocompatibility, immunogenicity, versatile surface bioengineering for desired biofunctionalities, capability of cargo loading and protection. We present OMV-based nanorobots designed for effective tumor therapy by leveraging the membrane properties of OMVs. These involve surface bioengineering of robotic body with cell-penetrating peptide for tumor targeting and penetration, which is further enhanced by active propulsion of nanorobots. Additionally, OMV nanorobots can effectively safeguard the loaded gene silencing tool, small interfering RNA (siRNA), from enzymatic degradation. Through systematic in vitro and in vivo studies using a rodent model, we demonstrate that these OMV nanorobots substantially enhanced siRNA delivery and immune stimulation, resulting in the utmost effectiveness in tumor suppression when juxtaposed with static groups, particularly evident in the orthotopic bladder tumor model. This OMV nanorobot opens an inspiring avenue to design advanced medical robots with expanded versatility and adaptability, broadening their operation scope in practical biomedical domains.

Guided diffusion for molecular generation with interaction prompt.

Molecular generative models have exhibited promising capabilities in designing molecules from scratch with high binding affinities in a predetermined protein pocket, offering potential synergies with traditional structural-based drug design strategy. However, the generative processes of such models are random and the atomic interaction information between ligand and protein are ignored. On the other hand, the ligand has high propensity to bind with residues called hotspots. Hotspot residues contribute to the majority of the binding free energies and have been recognized as appealing targets for designed molecules. In this work, we develop an interaction prompt guided diffusion model, InterDiff to deal with the challenges. Four kinds of atomic interactions are involved in our model and represented as learnable vector embeddings. These embeddings serve as conditions for individual residue to guide the molecular generative process. Comprehensive in silico experiments evince that our model could generate molecules with desired ligand-protein interactions in a guidable way. Furthermore, we validate InterDiff on two realistic protein-based therapeutic agents. Results show that InterDiff could generate molecules with better or similar binding mode compared to known targeted drugs.

A genomic and epigenomic atlas of prostate cancer in Asian populations.

Prostate cancer is the second most common cancer in men worldwide1. Over the past decade, large-scale integrative genomics efforts have enhanced our understanding of this disease by characterizing its genetic and epigenetic landscape in thousands of patients2,3. However, most tumours profiled in these studies were obtained from patients from Western populations. Here we produced and analysed whole-genome, whole-transcriptome and DNA methylation data for 208 pairs of tumour tissue samples and matched healthy control tissue from Chinese patients with primary prostate cancer. Systematic comparison with published data from 2,554 prostate tumours revealed that the genomic alteration signatures in Chinese patients were markedly distinct from those of Western cohorts: specifically, 41% of tumours contained mutations in FOXA1 and 18% each had deletions in ZNF292 and CHD1. Alterations of the genome and epigenome were correlated and were predictive of disease phenotype and progression. Coding and noncoding mutations, as well as epimutations, converged on pathways that are important for prostate cancer, providing insights into this devastating disease. These discoveries underscore the importance of including population context in constructing comprehensive genomic maps for disease.

CROST: a comprehensive repository of spatial transcriptomics.

The development of spatial transcriptome sequencing technology has revolutionized our comprehension of complex tissues and propelled life and health sciences into an era of spatial omics. However, the current availability of databases for accessing and analyzing spatial transcriptomic data is limited. In response, we have established CROST (https://ngdc.cncb.ac.cn/crost), a comprehensive repository of spatial transcriptomics. CROST encompasses high-quality samples and houses 182 spatial transcriptomic datasets from diverse species, organs, and diseases, comprising 1033 sub-datasets and 48 043 tumor-related spatially variable genes (SVGs). Additionally, it encompasses a standardized spatial transcriptome data processing pipeline, integrates single-cell RNA sequencing deconvolution spatial transcriptomics data, and evaluates correlation, colocalization, intercellular communication, and biological function annotation analyses. Moreover, CROST integrates the transcriptome, epigenome, and genome to explore tumor-associated SVGs and provides a comprehensive understanding of their roles in cancer progression and prognosis. Furthermore, CROST provides two online tools, single-sample gene set enrichment analysis and SpatialAP, for users to annotate and analyze the uploaded spatial transcriptomics data. The user-friendly interface of CROST facilitates browsing, searching, analyzing, visualizing, and downloading desired information. Collectively, CROST offers fresh and comprehensive insights into tissue structure and a foundation for understanding multiple biological mechanisms in diseases, particularly in tumor tissues.

TargetGene: a comprehensive database of cell-type-specific target genes for genetic variants.

Annotating genetic variants to their target genes is of great importance in unraveling the causal variants and genetic mechanisms that underlie complex diseases. However, disease-associated genetic variants are often located in non-coding regions and manifest context-specific effects, making it challenging to accurately identify the target genes and regulatory mechanisms. Here, we present TargetGene (https://ngdc.cncb.ac.cn/targetgene/), a comprehensive database reporting target genes for human genetic variants from various aspects. Specifically, we collected a comprehensive catalog of multi-omics data at the single-cell and bulk levels and from various human tissues, cell types and developmental stages. To facilitate the identification of Single Nucleotide Polymorphism (SNP)-to-gene connections, we have implemented multiple analytical tools based on chromatin co-accessibility, 3D interaction, enhancer activities and quantitative trait loci, among others. We applied the pipeline to evaluate variants from nearly 1300 Genome-wide association studies (GWAS) and assembled a comprehensive atlas of multiscale regulation of genetic variants. TargetGene is equipped with user-friendly web interfaces that enable intuitive searching, navigation and browsing through the results. Overall, TargetGene provides a unique resource to empower researchers to study the regulatory mechanisms of genetic variants in complex human traits.

HALL: a comprehensive database for human aging and longevity studies.

Diverse individuals age at different rates and display variable susceptibilities to tissue aging, functional decline and aging-related diseases. Centenarians, exemplifying extreme longevity, serve as models for healthy aging. The field of human aging and longevity research is rapidly advancing, garnering significant attention and accumulating substantial data in recent years. Omics technologies, encompassing phenomics, genomics, transcriptomics, proteomics, metabolomics and microbiomics, have provided multidimensional insights and revolutionized cohort-based investigations into human aging and longevity. Accumulated data, covering diverse cells, tissues and cohorts across the lifespan necessitates the establishment of an open and integrated database. Addressing this, we established the Human Aging and Longevity Landscape (HALL), a comprehensive multi-omics repository encompassing a diverse spectrum of human cohorts, spanning from young adults to centenarians. The core objective of HALL is to foster healthy aging by offering an extensive repository of information on biomarkers that gauge the trajectory of human aging. Moreover, the database facilitates the development of diagnostic tools for aging-related conditions and empowers targeted interventions to enhance longevity. HALL is publicly available at https://ngdc.cncb.ac.cn/hall/index.

TarKG: a comprehensive biomedical knowledge graph for target discovery.

MOTIVATION: Target discovery is a crucial step in drug development, as it directly affects the success rate of clinical trials. Knowledge graphs (KGs) offer unique advantages in processing complex biological data and inferring new relationships. Existing biomedical KGs primarily focus on tasks such as drug repositioning and drug-target interactions, leaving a gap in the construction of KGs tailored for target discovery. RESULTS: We established a comprehensive biomedical KG focusing on target discovery, termed TarKG, by integrating seven existing biomedical KGs, nine public databases, and traditional Chinese medicine knowledge databases. TarKG consists of 1 143 313 entities and 32 806 467 relations across 15 entity categories and 171 relation types, all centered around 3 core entity types: Disease, Gene, and Compound. TarKG provides specialized knowledges for the core entities including chemical structures, protein sequences, or text descriptions. By using different KG embedding algorithms, we assessed the knowledge completion capabilities of TarKG, particularly for disease-target link prediction. In case studies, we further examined TarKG's ability to predict potential protein targets for Alzheimer's disease (AD) and to identify diseases potentially associated with the metallo-deubiquitinase CSN5, using literature analysis for validation. Furthermore, we provided a user-friendly web server (https://tarkg.ddtmlab.org) that enables users to perform knowledge retrieval and relation inference using TarKG. AVAILABILITY AND IMPLEMENTATION: TarKG is accessible at https://tarkg.ddtmlab.org.

Gene Architectures that Minimize Cost of Gene Expression.

Gene expression burdens cells by consuming resources and energy. While numerous studies have investigated regulation of expression level, little is known about gene design elements that govern expression costs. Here, we ask how cells minimize production costs while maintaining a given protein expression level and whether there are gene architectures that optimize this process. We measured fitness of ∼14,000 E. coli strains, each expressing a reporter gene with a unique 5' architecture. By comparing cost-effective and ineffective architectures, we found that cost per protein molecule could be minimized by lowering transcription levels, regulating translation speeds, and utilizing amino acids that are cheap to synthesize and that are less hydrophobic. We then examined natural E. coli genes and found that highly expressed genes have evolved more forcefully to minimize costs associated with their expression. Our study thus elucidates gene design elements that improve the economy of protein expression in natural and heterologous systems.

MeDBA: the Metalloenzyme Data Bank and Analysis platform.

Metalloenzymes are attractive research targets in fields of chemistry, biology, and medicine. Given that metalloenzymes can manifest conservation of metal-coordination and ligand binding modes, the excavation and expansion of metalloenzyme-specific knowledge is of interest in bridging metalloenzyme-related fields. Building on our previous metalloenzyme-ligand association database, MeLAD, we have expanded the scope of metalloenzyme-specific knowledge and services, by forming a versatile platform, termed the Metalloenzyme Data Bank and Analysis (MeDBA). The MeDBA provides: (i) manual curation of metalloenzymes into different categories, that this M-I, M-II and M-III; (ii) comprehensive information on metalloenzyme activities, expression profiles, family and disease links; (iii) structural information on metalloenzymes, in particular metal binding modes; (iv) metalloenzyme substrates and bioactive molecules acting on metalloenzymes; (v) excavated metal-binding pharmacophores and (vi) analysis tools for structure/metal active site comparison and metalloenzyme profiling. The MeDBA is freely available at https://medba.ddtmlab.org.

ASCancer Atlas: a comprehensive knowledgebase of alternative splicing in human cancers.

Alternative splicing (AS) is a fundamental process that governs almost all aspects of cellular functions, and dysregulation in this process has been implicated in tumor initiation, progression and treatment resistance. With accumulating studies of carcinogenic mis-splicing in cancers, there is an urgent demand to integrate cancer-associated splicing changes to better understand their internal cross-talks and functional consequences from a global view. However, a resource of key functional AS events in human cancers is still lacking. To fill the gap, we developed ASCancer Atlas (https://ngdc.cncb.ac.cn/ascancer), a comprehensive knowledgebase of aberrant splicing in human cancers. Compared to extant databases, ASCancer Atlas features a high-confidence collection of 2006 cancer-associated splicing events experimentally proved to promote tumorigenesis, a systematic splicing regulatory network, and a suit of multi-scale online analysis tools. For each event, we manually curated the functional axis including upstream splicing regulators, splicing event annotations, downstream oncogenic effects, and possible therapeutic strategies. ASCancer Atlas also houses about 2 million computationally putative splicing events. Additionally, a user-friendly web interface was built to enable users to easily browse, search, visualize, analyze, and download all splicing events. Overall, ASCancer Atlas provides a unique resource to study the functional roles of splicing dysregulation in human cancers.

MethBank 4.0: an updated database of DNA methylation across a variety of species.

DNA methylation, as the most intensively studied epigenetic mark, regulates gene expression in numerous biological processes including development, aging, and disease. With the rapid accumulation of whole-genome bisulfite sequencing data, integrating, archiving, analyzing, and visualizing those data becomes critical. Since its first publication in 2015, MethBank has been continuously updated to include more DNA methylomes across more diverse species. Here, we present MethBank 4.0 (https://ngdc.cncb.ac.cn/methbank/), which reports an increase of 309% in data volume, with 1449 single-base resolution methylomes of 23 species, covering 236 tissues/cell lines and 15 biological contexts. Value-added information, such as more rigorous quality evaluation, more standardized metadata, and comprehensive downstream annotations have been integrated in the new version. Moreover, expert-curated knowledge modules of featured differentially methylated genes associated with biological contexts and methylation analysis tools have been incorporated as new components of MethBank. In addition, MethBank 4.0 is equipped with a series of new web interfaces to browse, search, and visualize DNA methylation profiles and related information. With all these improvements, we believe the updated MethBank 4.0 will serve as a fundamental resource to provide a wide range of data services for the global research community.

Optimising first-line subtyping-based therapy in triple-negative breast cancer (FUTURE-SUPER): a multi-cohort, randomised, phase 2 trial.

BACKGROUND: Triple-negative breast cancers display heterogeneity in molecular drivers and immune traits. We previously classified triple-negative breast cancers into four subtypes: luminal androgen receptor (LAR), immunomodulatory, basal-like immune-suppressed (BLIS), and mesenchymal-like (MES). Here, we aimed to evaluate the efficacy and safety of subtyping-based therapy in the first-line treatment of triple-negative breast cancer. METHODS: FUTURE-SUPER is an ongoing, open-label, randomised, controlled phase 2 trial being conducted at Fudan University Shanghai Cancer Center (FUSCC), Shanghai, China. Eligible participants were females aged 18-70 years, with an Eastern Cooperative Oncology Group performance status of 0-1, and histologically confirmed, untreated metastatic or recurrent triple-negative breast cancer. After categorising participants into five cohorts according to molecular subtype and genomic biomarkers, participants were randomly assigned (1:1) with a block size of 4, stratified by subtype, to receive, in 28-day cycles, nab-paclitaxel (100 mg/m2, intravenously on days 1, 8, and 15) alone (control group) or with a subtyping-based regimen (subtyping-based group): pyrotinib (400 mg orally daily) for the LAR-HER2mut subtype, everolimus (10 mg orally daily) for the LAR-PI3K/AKTmut and MES-PI3K/AKTmut subtypes, camrelizumab (200 mg intravenously on days 1 and 15) and famitinib (20 mg orally daily) for the immunomodulatory subtype, and bevacizumab (10 mg/kg intravenously on days 1 and 15) for the BLIS/MES-PI3K/AKTWT subtype. The primary endpoint was investigator-assessed progression-free survival for the pooled subtyping-based group versus the control group in the intention-to-treat population (all randomly assigned participants). Safety was analysed in all patients with safety records who received at least one dose of study drug. This study is registered with ClinicalTrials.gov (NCT04395989). FINDINGS: Between July 28, 2020, and Oct 16, 2022, 139 female participants were enrolled and randomly assigned to the subtyping-based group (n=69) or control group (n=70). At the data cutoff (May 31, 2023), the median follow-up was 22·5 months (IQR 15·2-29·0). Median progression-free survival was significantly longer in the pooled subtyping-based group (11·3 months [95% CI 8·6-15·2]) than in the control group (5·8 months [4·0-6·7]; hazard ratio 0·44 [95% CI 0·30-0·65]; p<0·0001). The most common grade 3-4 treatment-related adverse events were neutropenia (21 [30%] of 69 in the pooled subtyping-based group vs 16 [23%] of 70 in the control group), anaemia (five [7%] vs none), and increased alanine aminotransferase (four [6%] vs one [1%]). Treatment-related serious adverse events were reported for seven (10%) of 69 patients in the subtyping-based group and none in the control group. No treatment-related deaths were reported in either group. INTERPRETATION: These findings highlight the potential clinical benefits of using molecular subtype-based treatment optimisation in patients with triple-negative breast cancer, suggesting a path for further clinical investigation. Phase 3 randomised clinical trials assessing the efficacy of subtyping-based regimens are now underway. FUNDING: National Natural Science Foundation of China, Natural Science Foundation of Shanghai, Shanghai Hospital Development Center, and Jiangsu Hengrui Pharmaceuticals. TRANSLATION: For the Chinese translation of the abstract see Supplementary Materials section.

Mechanistic study of the transmission pattern of the SARS-CoV-2 omicron variant.

The omicron variant of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) characterized by 30 mutations in its spike protein, has rapidly spread worldwide since November 2021, significantly exacerbating the ongoing COVID-19 pandemic. In order to investigate the relationship between these mutations and the variant's high transmissibility, we conducted a systematic analysis of the mutational effect on spike-angiotensin-converting enzyme-2 (ACE2) interactions and explored the structural/energy correlation of key mutations, utilizing a reliable coarse-grained model. Our study extended beyond the receptor-binding domain (RBD) of spike trimer through comprehensive modeling of the full-length spike trimer rather than just the RBD. Our free-energy calculation revealed that the enhanced binding affinity between the spike protein and the ACE2 receptor is correlated with the increased structural stability of the isolated spike protein, thus explaining the omicron variant's heightened transmissibility. The conclusion was supported by our experimental analyses involving the expression and purification of the full-length spike trimer. Furthermore, the energy decomposition analysis established those electrostatic interactions make major contributions to this effect. We categorized the mutations into four groups and established an analytical framework that can be employed in studying future mutations. Additionally, our calculations rationalized the reduced affinity of the omicron variant towards most available therapeutic neutralizing antibodies, when compared with the wild type. By providing concrete experimental data and offering a solid explanation, this study contributes to a better understanding of the relationship between theories and observations and lays the foundation for future investigations.

Predicting Mutational Effects on Ca2+-Activated Chloride Conduction of TMEM16A Based on a Simulation Study.

The dysfunction and defects of ion channels are associated with many human diseases, especially for loss-of-function mutations in ion channels such as cystic fibrosis transmembrane conductance regulator mutations in cystic fibrosis. Understanding ion channels is of great current importance for both medical and fundamental purposes. Such an understanding should include the ability to predict mutational effects and describe functional and mechanistic effects. In this work, we introduce an approach to predict mutational effects based on kinetic information (including reaction barriers and transition state locations) obtained by studying the working mechanism of target proteins. Specifically, we take the Ca2+-activated chloride channel TMEM16A as an example and utilize the computational biology model to predict the mutational effects of key residues. Encouragingly, we verified our predictions through electrophysiological experiments, demonstrating a 94% prediction accuracy regarding mutational directions. The mutational strength assessed by Pearson's correlation coefficient is -0.80 between our calculations and the experimental results. These findings suggest that the proposed methodology is reliable and can provide valuable guidance for revealing functional mechanisms and identifying key residues of the TMEM16A channel. The proposed approach can be extended to a broad scope of biophysical systems.

Enhanced Recognition of a Herbal Compound Epiberberine by a DNA Quadruplex-Duplex Structure.

The small molecule epiberberine (EPI) is a natural alkaloid with versatile bioactivities against several diseases including cancer and bacterial infection. EPI can induce the formation of a unique binding pocket at the 5' side of a human telomeric G-quadruplex (HTG) sequence with four telomeric repeats (Q4), resulting in a nanomolar binding affinity (KD approximately 26 nM) with significant fluorescence enhancement upon binding. It is important to understand (1) how EPI binding affects HTG structural stability and (2) how enhanced EPI binding may be achieved through the engineering of the DNA binding pocket. In this work, the EPI-binding-induced HTG structure stabilization effect was probed by a peptide nucleic acid (PNA) invasion assay in combination with a series of biophysical techniques. We show that the PNA invasion-based method may be useful for the characterization of compounds binding to DNA (and RNA) structures under physiological conditions without the need to vary the solution temperature or buffer components, which are typically needed for structural stability characterization. Importantly, the combination of theoretical modeling and experimental quantification allows us to successfully engineer Q4 derivative Q4-ds-A by a simple extension of a duplex structure to Q4 at the 5' end. Q4-ds-A is an excellent EPI binder with a KD of 8 nM, with the binding enhancement achieved through the preformation of a binding pocket and a reduced dissociation rate. The tight binding of Q4 and Q4-ds-A with EPI allows us to develop a novel magnetic bead-based affinity purification system to effectively extract EPI from Rhizoma coptidis (Huang Lian) extracts.

Diverse cell death patterns-related signature for predicting prognosis and drug sensitivity of osteosarcoma patients.

BACKGROUND: Programmed cell death (PCD) is a natural process in which cells undergo controlled self-destruction, which plays a crucial role in maintaining tissue homeostasis and eliminating damaged or unnecessary cells. The connection between PCD and osteosarcoma was explored in the present study. METHODS: Twelve types of PCD were collected for developing a prognostic signature in osteosarcoma using machine learning algorithms. The prognostic value, pathway annotation and drug prediction of the signature were explored. RESULTS: Telomerase reverse transcriptase (TERT) was found to be a potent hazardous marker in osteosarcoma and could facilitate the proliferation and migration of osteosarcoma. CONCLUSIONS: In summary, the present study has developed a prognostic signature for osteosarcoma and identifies TERT as a potent hazardous gene. The study suggests that further research is needed to address the underlying mechanism of how TERT affects the immune response in osteosarcoma.

Antigen processing and presentation-related signature-derived BNIP3 is a novel oncogene and immunotherapy determinant in osteosarcoma based on machine learning and in vitro validation.

BACKGROUND: In recent decades, osteosarcoma has remained the most prevalent kind of malignant tumor. An important and crucial factor in immunotherapy is antigen processing and presentation (APP). The specific functions and pathogenic processes of APP in osteosarcoma have not, however, been studied. METHODS: Patients with osteosarcoma were divided into groups using APP-related genes. Machine learning was used to further build the APP-related score. Investigated in-depth were the prognostic relevance of the score, mutation features, immunological aspects, and pharmacological prediction performance. Investigations of the prognostic utility, immunological traits, drug prediction effectiveness and immunotherapy prediction of BNIP3 were performed in-depth. RESULTS: Investigations by cell counting kit-8, Transwell and 5-ethynyl-2-deoxyuridine (EdU) demonstrated that BNIP3 is an osteosarcoma tumor accelerator. The osteosarcoma gene BNIP3 may promote macrophage migration. The APP-related score shows potential for clinical use. CONCLUSIONS: It was anticipated that more in vitro and in vivo studies would confirm BNIP3's tumorigenic and immunogenic processes in osteosarcoma.

Disulfidptosis-related PABPC3 promotes tumor progression and inhibits immune activity in osteosarcoma.

BACKGROUND: Osteosarcoma is a very aggressive bone tumor mainly affecting teens and young adults. Disulfidptosis is a metabolic-related form of regulated cell death. However, the interconnection between disulfidptosis and osteosarcoma has not been explored. METHODS: In the present study, disulfidptosis-related clusters were identified in osteosarcoma using the nonnegative matrix factorization clustering method. PABPC3 was identified as a hazardous gene in osteosarcoma using machine learning algorithms, CoxBoost, and Random Survival Forest. The prognostic value, pathway annotation, immune characteristics, and drug prediction of PABPC3 were systematically explored. MTT (i.e., 3-(4, 5-dimethyl thiazol-2-yl)-2,5-diphenytetrazolium bromide), EdU (ie. 5-ethyny-2'-deoxvuridine), and Transwell assays were used for in vitro validation of PABPC3. RESULTS: The disulfidptosis-related clusters could distinguish survival outcomes of osteosarcoma patients. PABPC3 could predict survival outcomes, immune activity, and drug response in osteosarcoma patients. Besides, PABPC3 was proven to facilitate the proliferation and migration of osteosarcoma. CONCLUSIONS: The present study is expected to establish the bridge between disulfidptosis and osteosarcoma. PABPC3 is expected to be further explored as a therapeutic target in osteosarcoma.

Engineered Microrobots for Targeted Delivery of Bacterial Outer Membrane Vesicles (OMV) in Thrombus Therapy.

In the realm of thrombosis treatment, bioengineered outer membrane vesicles (OMVs) offer a novel and promising approach, as they have rich content of bacterial-derived components. This study centers on OMVs derived from Escherichia coli BL21 cells, innovatively engineered to encapsulate the staphylokinase-hirudin fusion protein (SFH). SFH synergizes the properties of staphylokinase (SAK) and hirudin (HV) to enhance thrombolytic efficiency while reducing the risks associated with re-embolization and bleeding. Building on this foundation, this study introduces two cutting-edge microrobotic platforms: SFH-OMV@H for venous thromboembolism (VTE) treatment, and SFH-OMV@MΦ, designed specifically for cerebral venous sinus thrombosis (CVST) therapy. These platforms have demonstrated significant efficacy in dissolving thrombi, with SFH-OMV@H showcasing precise vascular navigation and SFH-OMV@MΦ effectively targeting cerebral thrombi. The study shows that the integration of these bioengineered OMVs and microrobotic systems marks a significant advancement in thrombosis treatment, underlining their potential to revolutionize personalized medical approaches to complex health conditions.

Enhanced Artificial Intelligence Strategies in Renal Oncology: Iterative Optimization and Comparative Analysis of GPT 3.5 Versus 4.0.

BACKGROUND: The rise of artificial intelligence (AI) in medicine has revealed the potential of ChatGPT as a pivotal tool in medical diagnosis and treatment. This study assesses the efficacy of ChatGPT versions 3.5 and 4.0 in addressing renal cell carcinoma (RCC) clinical inquiries. Notably, fine-tuning and iterative optimization of the model corrected ChatGPT's limitations in this area. METHODS: In our study, 80 RCC-related clinical questions from urology experts were posed three times to both ChatGPT 3.5 and ChatGPT 4.0, seeking binary (yes/no) responses. We then statistically analyzed the answers. Finally, we fine-tuned the GPT-3.5 Turbo model using these questions, and assessed its training outcomes. RESULTS: We found that the average accuracy rates of answers provided by ChatGPT versions 3.5 and 4.0 were 67.08% and 77.50%, respectively. ChatGPT 4.0 outperformed ChatGPT 3.5, with a higher accuracy rate in responses (p < 0.05). By counting the number of correct responses to the 80 questions, we then found that although ChatGPT 4.0 performed better (p < 0.05), both versions were subject to instability in answering. Finally, by fine-tuning the GPT-3.5 Turbo model, we found that the correct rate of responses to these questions could be stabilized at 93.75%. Iterative optimization of the model can result in 100% response accuracy. CONCLUSION: We compared ChatGPT versions 3.5 and 4.0 in addressing clinical RCC questions, identifying their limitations. By applying the GPT-3.5 Turbo fine-tuned model iterative training method, we enhanced AI strategies in renal oncology. This approach is set to enhance ChatGPT's database and clinical guidance capabilities, optimizing AI in this field.