The Li Man Team recommends:

2025 marks the inaugural year of virtual cells. As AI technology accelerates its development, the creation of virtual cells isn’t just a technological breakthrough—it also represents a profound advancement in our understanding of life sciences. The Grainman high-throughput cell-editing platform is set to empower the future of virtual cells by providing more high-quality samples and robust validation solutions!

The following article is sourced from Guomics, authored by Guomics.

 

Westlake Symposium on AI Proteomics and Virtual Cell

 

From October 8 to 9, 2025, the event will be hosted by the School of Medicine at Westlake University. Westlake Symposium on AI-driven Proteomics and Virtual Cells It was successfully held in Hangzhou. The conference brought together more than 30 leading scholars from China and abroad to jointly explore AI-Powered Cutting-Edge Breakthroughs in Life Sciences The conference highlights focused on two cutting-edge themes:

AI Proteomics: Innovative applications of deep learning and machine learning in mass spectrometry data analysis, protein identification, and functional prediction—including the establishment of the AI Proteomics Competition (AIPC), an international algorithmic contest.

AI Virtual Cells: Recent advances in the integration of multi-omics time- and spatially resolved data with cell modeling are driving the digital analysis and prediction of life systems.

Researchers from nearly 300 institutions—including prestigious universities such as Tsinghua University, Peking University, Fudan University, Shanghai Jiao Tong University, Zhejiang University, University of Science and Technology of China, Nanjing University, Sun Yat-sen University, Wuhan University, Nankai University, Tianjin University, the University of Hong Kong, the Chinese University of Hong Kong, the Hong Kong University of Science and Technology, Imperial College London, Oxford University, Harvard Medical School, Hong Kong Polytechnic University, Southern University of Science and Technology, Xi'an Jiaotong University, East China University of Science and Technology, China Agricultural University, China Pharmaceutical University, Peking Union Medical College, Capital Medical University, Southern Medical University, Southeast University, Sichuan University, Shandong University, Chongqing University, Xiamen University, Hunan University, Zhejiang University of Technology, Zhejiang University of Traditional Chinese Medicine, Fujian University of Traditional Chinese Medicine, Zhejiang University of Science and Technology, Zhejiang A&F University, Yangzhou University, Wannan Medical College, Qingdao University of Science and Technology, Shenyang Pharmaceutical University—as well as representatives from research and medical institutions like the Academy of Military Medical Sciences, Beijing Institute of Life Sciences, Shanghai Institute of Materia Medica, Chinese Academy of Sciences, Institute of Genetics and Developmental Biology, Institute of Zoology, Dalian Institute of Chemical Physics, National Center for Protein Sciences (Beijing), Hangzhou Institute of Medical Research, Chinese Academy of Sciences, CAS-Hangzhou Institute for Advanced Study, Pengcheng Laboratory, Westlake Laboratory, Zhejiang Academy of Agricultural Sciences, Peking University Third Hospital, Hunan Provincial Children's Hospital, Ruijin Hospital, Hangzhou First People's Hospital, Chongqing People's Hospital, Zhongshan Ophthalmic Center at Sun Yat-sen University, Zhejiang Provincial Tumor Hospital, Shenzhen Third People's Hospital, Shanghai Sixth People's Hospital, Quadram Institute, Ruijin Hospital affiliated to Shanghai Jiao Tong University School of Medicine, and PICB—came together to participate in this conference. Representatives from companies including L'Oréal, Huawei, Nanomics Biotech, Absea, and Liman Bio, along with several media outlets, also attended the event.

Attendees at the meeting gathered around AI Proteomics And AI Virtual Cells Two cutting-edge topics were brilliantly presented, covering fields such as biology, artificial intelligence, and data science, effectively advancing the deep integration and convergence of AI with life sciences. Attendees engaged enthusiastically in lively discussions with the speakers, creating a vibrant academic atmosphere at the event.

 

✨ Opening

President Yigong Shi

West Lake University

The conference opened with President Yi-Gong Shi first extending a warm welcome to all the distinguished guests, followed by a brief introduction to Westlake University. Located in the picturesque city of Hangzhou, Westlake University stands as China’s first new-type research-oriented university in modern history. Leveraging a unique dual-support mechanism—combining public and private funding—Westlake University is at the forefront of higher education reform in China. Originally established in 2016 as the Westlake Institute for Advanced Study, the university has consistently been committed to cultivating top-tier talent, driving breakthroughs in both fundamental research and cutting-edge technological innovation, and harnessing science and technology to propel human progress. He also highlighted Hangzhou’s remarkable advancements in various fields of artificial intelligence, particularly its outstanding achievements in life sciences.

01 Session 1: Proteomics

Professor Guo Tiannan

West Lake University

AI proteomics, virtual cells, and the AI Proteomics Competition (AIPC)

Professor Guo Tiannan outlined the latest advancements of the Artificial Intelligence Virtual Cell (AIVC) project. He systematically reviewed key international collaborations, major breakthroughs in spatiotemporal proteomics, and the WAY project, highlighting both the profound significance of the virtual cell vision and the challenges that remain in realizing this ambitious goal. He introduced the AIVC framework’s three core pillars—prior knowledge, static structure, and dynamic state—all of which are underpinned by closed-loop active learning technology. Building such virtual cells will continue to rely on cutting-edge AI capabilities. Just as modern generative AI can restore blurry or incomplete images into clear, full-fledged visuals—and even transform static pictures into dynamic videos—AIVC aims to reconstruct the comprehensive, dynamic behaviors of cells from fragmented omics data. His team has already amassed a large-scale spatiotemporal proteomics dataset and, leveraging tissue expansion technology, independently developed the FAXP platform, achieving world-class precision in spatial proteomics. This innovative technique physically expands tissues—effectively increasing the volume of individual cells by approximately 100 times—enabling the quantitative identification of more than 3,500 proteins within single cells. Additionally, the team has proposed the PMMP paradigm (Perturbation—Measurement—Modeling—Prediction) as a systematic workflow for predictive biology, providing a proof-of-concept approach to constructing virtual cells centered around proteomics. Building on this framework, the team has generated extensive time-resolved perturbation proteomics datasets—including ProteinTalks v1.0 through v4.0—that cover pan-cancer cell lines treated with thousands of compounds. Finally, in their forward-looking WAY paper, they suggest using "virtual yeast" as a practical blueprint. Before scaling the AIVC concept to human systems, this approach will first serve as an experimental testbed to validate the scalability and feasibility of the idea.

Prof. Ruedi Aebersold

ETH Zurich

The adaptable, modular proteome defines AIVC

In this report, Professor Ruedi Aebersold discusses the π-HuB project—a groundbreaking initiative aimed at digitally mapping the "spatial" proteome at both tissue and individual levels. The development of π-HuB Human Navigation will significantly advance our mechanistic understanding and predictive capabilities in disease research. Addressing the current high failure rate in clinical trials, the report emphasizes that the core focus should shift toward exploring biological mechanisms rather than relying solely on pattern recognition. To achieve this, the concept of AIVC (AI-Enabled Virtual Cells)—modeled after酿酒er's yeast—is proposed. By integrating genomic and proteomic data, AIVC enables the high-density acquisition of multi-omics information within the yeast system, capturing a comprehensive array of details including proteins, protein isoforms, post-translational modifications, protein interactions, structural features, functional activities, and even RNA-protein correlations.

Cells are "complex adaptive systems," involving the coordinated regulation of multi-component states, local interactions, and emergent behaviors. The case study demonstrates the synchronized fluctuations between the proteome and phosphoproteome during meiosis, as well as the dynamic mapping of variations spanning from the transcriptome to the proteome and finally to the phosphoproteome. At molecular levels closer to observable phenotypes, signals associated with physiological traits become more readily detectable. Based on these insights, we propose the following recommendations: conduct systematic perturbations and time-series measurements on the same sample; centralize or rigorously coordinate data collection efforts; and systematically identify and iteratively refine the most critical data types for mechanistic modeling. Yeast will serve as the AIVC experimental platform, and the methodological lessons learned from this model can be extended to π-HuB, ultimately enhancing our ability to precisely predict and prevent human diseases.

Prof. Matthias Mann

Max Planck Institute of Biochemistry

Proteomics data generation strategies for robust AI models in biomedicine

Professor Matthias Mann’s presentation showcased his lab’s cutting-edge work in developing proteomics technologies and their biomedical applications: Highlights included the AI-driven peptide prediction tool AlphaPeptDeep, AlphaNovo, which enables "deep learning-based decoding" that directly translates spectra into sequences, the use of large-scale datasets to train advanced models, and the integration with the Evosep-Astral platform to achieve high-throughput protein profiling of up to 500 samples per day. The team also demonstrated groundbreaking innovations such as PELSA chemical proteomics, comprehensive maps of virus-human interaction networks, spatial single-cell proteomics technology, and the ADAPT-MS machine learning diagnostic framework—each underscoring the transformative potential of AI and proteomics in advancing virtual cell modeling and paving the way for precision medicine.

Prof. Albert Heck

Utrecht University

Tackling a new frontier in proteomics: de novo sequencing of endogenous antibodies

Professor Albert J. R. Heck from Utrecht University presented his team's cutting-edge work in the field of proteomics: tackling the next frontier in proteomics—how much do we really know about antibodies? Specifically, they are pioneering de novo sequencing of endogenous antibodies using mass spectrometry technology. This innovative approach enables direct analysis of the unique antibody repertoires generated by individuals in response to pathogens, paving the way for identifying the most promising candidate molecules for biotherapeutic applications. Immunoglobulins are among the most abundant proteins found both in the human body and in blood, playing a critical role in the humoral immune system by safeguarding us against microbial infections. By capturing IgG1 derived from blood samples and employing advanced techniques such as EAciD fragmentation along with the novel HTA protease, the team has achieved high-quality, high-coverage antibody sequence analysis. This method has already been successfully applied to identify highly effective IgG antibodies targeting emerging variants—such as JN.1—from plasma samples of COVID-19 patients. These groundbreaking results demonstrate that this approach represents a revolutionary strategy for rapidly uncovering potent therapeutic antibodies directly from patient samples, a goal that Professor Heck’s company, Abvion, is passionately committed to realizing.

Prof. Uwe Völker

University of Greifswald

Learning from the genomics community – A pathway to large-scale collaboration projects in proteomics

In his report, Professor Uwe Völker systematically outlined the advantages, advancements, and challenges of population-based omics research. He first highlighted that this field already boasts clearly defined, highly standardized workflows, robust international collaboration networks, and increasingly affordable high-throughput technologies—all of which are driving forces behind its rapid progress. Using large-scale cohorts such as SHIP and the UK Biobank as examples, he demonstrated how these studies, by integrating multi-omics data, have profoundly unveiled the biological underpinnings of health and disease. The professor particularly emphasized recent breakthroughs in plasma proteomics, noting that the remarkable improvements in throughput and coverage now enable large-scale cohort analyses while uncovering a wealth of protein molecules linked to genetics and health. Citing cross-laboratory validation studies, he underscored that even with varying workflows, excellent data reproducibility and comparability can be achieved through well-designed standardization strategies. At the same time, he candidly acknowledged the current challenges, including issues related to data sharing and privacy protection, the need for standardized sample pre-processing protocols, and the ongoing lack of representative reference samples. Finally, looking ahead, the professor stressed that, in the era of "big science," strengthening international collaboration via platforms like HUPO—and establishing standardized frameworks and reference systems—will be crucial for achieving seamless data integration, ultimately paving the way for early disease detection and personalized medicine.

02 Session 2: Biomedicine

Prof. Charles Boone

University of Toronto

A global genetic interaction map of a human cell reveals conserved principles of genetic networks.

Professor Charles Boone’s team systematically elucidated the profound conservation of genetic regulatory networks—from yeast to human cells. The team first leveraged high-throughput technologies to construct a comprehensive yeast genetic interaction network (CellMap), and groundbreakingly integrated multidimensional data using a deep-learning model (BIONIC), significantly enhancing both the network’s coverage and its ability to predict functional relationships. They successfully applied this strategy to human HAP1 cells, creating the first large-scale map of human gene interactions. Their findings revealed that the networks in humans and yeast share remarkable conservation at both the level of overarching organizational principles and specific interaction pathways—such as ECM9/PTAR1—providing a powerful example and invaluable resource for systematically dissecting human gene functions and disease mechanisms using model organisms.

Professor Li Ming

Central China Institute of Artificial Intelligence

University of Waterloo

Decoding the Human Immunopeptidome with AI

Professor Ming Li from the University of Waterloo in Canada and the Central Plains Institute of Artificial Intelligence Industry Technology delivered a presentation titled "Decoding the Human Immunopeptidome with Artificial Intelligence," unveiling a novel strategy that leverages mass spectrometry combined with deep learning to comprehensively analyze the immunopeptidome. Professor Li highlighted that immune system-related diseases, such as cancer and autoimmune disorders, are closely tied to imbalances in immune recognition. He emphasized that building a comprehensive immunopeptidome database is crucial for advancing personalized immunotherapy. The professor also showcased his team's groundbreaking achievements in areas like cross-linking removal from FFPE samples, identification of non-classical immune peptides, PTM characterization, AI-driven peptide search algorithms, and TCR-pMHC binding prediction. Furthermore, he outlined an ambitious plan to systematically analyze the immunopeptidome using large-scale AI models and establish a public database, laying a robust foundation for precision immunotherapy.

Prof. Brenda Andrews

University of Toronto

Single-cell imaging to study proteome dynamics in yeast

Professor Brenda Andrews introduced a high-throughput "phenomics" research platform that integrates yeast’s synthetic genetic array (SGA) technology with automated single-cell imaging, enabling a systematic investigation of the dynamic collapse process in cellular structures. The study reveals that when essential gene functions are disrupted, the resulting collapse of cellular architecture is not a random event—but rather unfolds through rapid, cascading effects, leading to an exponentially accelerating breakdown. Importantly, the trajectory of this structural collapse closely mirrors the initial biological process that was perturbed; for instance, inhibiting vesicle transport triggers an acute and widespread loss of cellular integrity. Remarkably, both the extent and speed of this structural disintegration can efficiently predict whether the cell will ultimately succumb to death. Furthermore, the model highlights that cells undergoing natural aging also experience a similar exponential collapse toward the end of their lifespan—often initiated by early defects in mitochondrial function.

Prof. Joseph Schacherer

University of Strasbourg

A deep exploration of the genotype-phenotype relationship through the lens of 1,086 near telomere-to-telomere yeast genomes

Professor Joseph Schacherer’s team has deeply unveiled the intricate relationship between genotype and phenotype by conducting near-telomere-to-telomere (T2T) whole-genome sequencing on 1,086 yeast strains. The team constructed a comprehensive genetic variation map of yeast, covering SNPs, InDels, and structural variations (SVs). By integrating multi-dimensional phenotypic data—such as transcriptomic and proteomic profiles—the researchers found that incorporating SVs and InDels could boost the model’s explained heritability by up to 15%. Their analysis revealed that structural variations not only exhibit more frequent associations with traits and greater pleiotropy but also play a more central role in regulating complex organism-level traits (e.g., growth rate), often surpassing molecular-level traits in importance. Moreover, the genetic architecture underlying these variations tends to follow a "polygenic, small-effect" pattern. This groundbreaking work underscores the critical value of structural variations in unraveling the "missing heritability" puzzle associated with complex traits, offering fresh insights for research in this field.

Prof. Connie Jimenez

Amsterdam University Medical Center

Transformer-based deep learning for next-generation mass spectrometry–based phosphoproteomics

Professor Connie Jimenez has been founding and leading the OncoProteomics Laboratory at Amsterdam University Medical Center since 2006, dedicating her research to integrating mass spectrometry-based proteomics with AI. Her report showcases how Transformer-based deep learning models can enhance the data analysis capabilities of DIA-MS phosphoproteomics, including the development and optimization of retention time, MS/MS, and ion mobility prediction models. The team has trained these models on large-scale datasets and recently released the open-source packages aiproteomics and iq 2.0, significantly boosting both the speed and accuracy of protein quantification algorithms. These groundbreaking findings pave the way for large-scale, precision analysis of tumor signaling pathways, offering innovative approaches to personalized cancer treatment.

Edited by Hou Jingyi

EMBO Press

Behind the scenes at EMBO Press

Hou Jingyi is under EMBO Press EMBO Molecular Medicine With Molecular Systems Biology A senior scientific editor, she highlighted the two journals' strong emphasis on systems biology, computational modeling, AI, and translational medicine, while also providing a detailed overview of their highly efficient, author-friendly services—such as "First-to-Protect," transferable peer review, and a streamlined rapid-publishing process. The presentation featured several notable papers, including single-cell proteomics research and the perturbation-prediction AI model "PerturbNet," underscoring EMBO Press's pivotal role in advancing cutting-edge fields like AI-driven proteomics and virtual cell science. EMBO Press’s platform is actively fostering innovation and collaboration within these dynamic areas.

Prof. Bernd Wollscheid

ETH Zurich

Virtual reality: how ML/AI-based strategies can shed light on the functional roles of surfaceome protein communities

Professor Bernd Wollscheid’s presentation explored how machine learning and artificial intelligence can be leveraged to decipher the functional architecture of the surfaceome—the dynamic community of proteins on the cell surface, over 80% of which remain largely uncharted. The talk highlighted technologies such as Cell Surface Capture (CSC) and LUX-MS, which enable the mapping of surface proteins and their nanoscale interactions with unprecedented precision. A machine-learning tool called SURFY has been developed to predict surfaceome proteins, while LUX-MS facilitates light-controlled, time-resolved analysis of protein complexes at the nanoscale. These approaches hold promising applications, including the analysis of cancers like lymphoma and lung cancer, insights into neuronal development, and the dissection of synapses triggered by immunotherapies. Ultimately, the goal is to create a virtual "Google Street View map" of the surfaceome—enabling the identification of novel therapeutic targets—and to bring this vision to life through collaborative efforts and the integration of large-scale proteomic datasets.

Prof. Ben Collins

Queen's University Belfast

Chemoproteomics in drug discovery – opportunities for AI?

Professor Ben Collins presented the application of chemical proteomics in drug discovery, as well as the opportunities for integrating AI into this field. He highlighted two emerging drug modalities—PROTACs (proteolysis-targeting chimeras) and covalent ligands—which hold promise for tackling the challenges posed by traditionally "undruggable" proteins. The Collins lab has developed high-throughput sample preparation and data-acquisition workflows, enabling time- and concentration-dependent assessments of protein degradation. He also showcased dual global proteomics and activity probe technologies, allowing simultaneous measurement of protein abundance and covalent ligand binding within a single experiment. In terms of AI applications, he explored the potential of protein-ligand structure-prediction models such as AlphaFold3 and Boltz-2, suggesting that chemical proteomics datasets could be leveraged to train affinity-prediction models. These models would capture biology-state-dependent ligand-binding dynamics, providing richer training data to power AI-driven drug discovery efforts.

Professor Wang Dong

Chengdu University of Traditional Chinese Medicine

Generating large-scale perturbation-induced transcriptome data for drug discovery

Professor Wang Dong's report focused on leveraging high-throughput transcriptomics technology to advance drug discovery. He introduced HTS² and the next-generation HiMAP-seq technology developed by his team—both methods capable of simultaneously analyzing the expression of thousands of genes across thousands of samples in a single assay, boasting high sensitivity, excellent reproducibility, and minimal cross-contamination. Building on this, they have constructed the CIGS database, which currently contains over 13,000 compounds and approximately 320 million gene expression events recorded in two distinct cell lines. This resource has already demonstrated its utility in identifying novel BRD4 inhibitors, such as luteolin, as well as anti-ferroptosis compounds like 2,4-dihydroxybenzaldehyde. Ultimately, this comprehensive dataset provides a robust foundation for drug development targeting diseases with unclear mechanisms, supports research into bioactive components of traditional Chinese medicine, and fuels AI-driven approaches to pharmaceutical innovation.

Professor Yue Jiaxing

Sun Yat-sen University

Universal telomere sequencing reveals hidden diversity underlying genome instability, aging, and cancer

In this report, Professor Yue Jiaxing introduced how telomere instability plays a critical role in the aging process and the development of cancer. However, existing sequencing methods have inherent limitations when it comes to detecting telomere regions. To address this, Professor Yue Jiaxing proposed a versatile, telomere-targeted sequencing approach called Termin-Seq, which is applicable to all eukaryotic organisms and has demonstrated exceptional sequencing performance across a variety of species, including humans, yeast, and mice.

Additionally, Termin-Seq can effectively capture telomere-related genetic perturbations. When applied to aging studies in mice, it revealed that telomere length significantly shortens with age. Further research using cancer cell lines uncovered genomic instability in telomeres, which was closely linked to tumor drug resistance. Notably, the addition of a telomerase inhibitor markedly alleviated the resistance issue associated with osimertinib in lung cancer treatment.

Dr. Liu Cui

SCIEX

Ultra-sensitive quantitative proteomic profiling of single or few cells enabled by the ZenoTOF 8600 ZT Scan DIA.

Dr. Liu Cui highlighted the high-performance SCIEX ZenoTOF mass spectrometry system, a platform that delivers a proven solution for spatial proteomics. The latest-generation ZenoTOF 8600, equipped with ZT Scan 2.0 technology, enables ultra-sensitive proteomic analysis down to the single-cell and even trace-cell levels.

Prof. Edouard Nice

Monash University

AI and the Path to Personalized/Precision Medicine

Professor Ed Nice summarized the rapid advancements in omics technologies in recent years, noting that the completion rates of the human genome and proteome have already surpassed 90%. Notably, the application of AI in personalized medicine has garnered significant attention, fostering collaborative efforts among interdisciplinary international teams. He discussed how AI is poised to reshape the field of pathology, as well as the critical responsibilities that pathologists must embrace moving forward. Additionally, he highlighted the market potential of personalized medicine, emerging trends in proteomics, and the challenges that may arise in leveraging artificial intelligence to achieve precision healthcare.

03

Session 3: AI

Prof. Yasset Perez-Riverol

EMBL's European Bioinformatics Institute

Quantum ecosystem: data, formats, and algorithms to generate AI-ready proteomics datasets

Professor Yasset Perez-Riverol systematically outlined the construction goals and key technological roadmap of the Quantms ecosystem. He pointed out that current proteomics data still exhibit significant variations in terms of scale, format, and quality control, severely limiting their usability in AI models. To address this challenge, Quantms is dedicated to efficiently transforming raw mass spectrometry data into "AI-ready" structured inputs through standardized data frameworks, scalable algorithmic modules, and fully automated processing workflows. Professor Yasset Perez-Riverol provided a detailed overview of how the ecosystem optimizes strategies across core stages such as data cleaning, retention time calibration, feature extraction, and quantitative consistency control, while emphasizing the critical role of format standardization in enhancing data reusability and fostering cross-project versatility. Additionally, Quantms seamlessly integrates with a variety of downstream machine learning frameworks, offering proteomics researchers a highly flexible and reproducible analysis platform. This presentation provides a vital technical pathway and practical reference for the deep integration of AI and proteomics research.

Prof. Chris Sander

Harvard Medical School

Machine learning for proteomic perturbation biology

Professor Chris Sander outlined the vision of "perturbation biology": by systematically perturbing cells and measuring their multidimensional responses, researchers aim to build computable, predictive models of virtual cells. He noted that the challenge lies in moving beyond single-protein measurements, integrating temporal data with domain-specific knowledge, and redefining cellular processes in a data-driven manner. His team is leveraging AI approaches—such as protein structure prediction and the EVcouplings design model—to guide enzyme engineering and drug discovery, including applications like enhancing target flexibility and optimizing combination therapies. Moreover, they’re expanding their scope to disease prevention, for instance, using electronic health records to predict pancreatic cancer risk. Ultimately, the goal is to develop multi-scale, executable models spanning from cells and tissues to organs, enabling precision medicine that seamlessly transitions from prediction to intervention.

Edited by Arunima Singh

Nature Methods

Publishing in Nature Methods and pursuing an editorial career

Nature Methods Dr. Arunima Singh, a senior editor of the journal, delivered an insightful presentation at the conference, detailing the journal’s key areas of focus, the composition of its editorial team, and the types of articles it publishes. She further shared the critical factors editors pay close attention to during the peer-review process—such as topic selection, novelty and significance, practical value and generality, validation, and applicability—and systematically explained the writing techniques specific to Methods-style articles.

Dr. Singh particularly emphasized, Nature Methods The core focus is on the method itself—authors should "make the method the star." Additionally, she points out that researchers should situate their work within a broader disciplinary context, not only outlining the current state-of-the-art but also emphasizing how their approach goes beyond existing technologies. At the same time, they should provide direct comparisons and demonstrate the method's practical applications. She also reminds us that the method's reproducibility and its accessibility—for both editors and reviewers—are equally critical.

Additionally, Dr. Singh shared insights into the daily responsibilities of a professional editor, as well as the experiences and skills required to become an editor for a scientific journal—such as honing writing and reading abilities, staying abreast of cutting-edge research trends, and actively building networks with editors. Her presentation provided valuable guidance for those aspiring to submit their work in the future. Nature Methods Researchers aspiring to work in scientific publishing also provided valuable guidance and inspiration.

Prof. Henning Hermjakob

European Bioinformatics Institute, EMBL-EBI

Reactome 4: Pathways Reimagined – Dynamic Visualization and Intelligent Chat

Professor Henning Hermjakob introduced the key updates in Reactome Database Version 4, highlighting the newly launched dynamic pathway diagram visualization feature and the integrated intelligent dialogue assistant, both designed to enhance user interaction and boost analytical efficiency. The latest version of Reactome enables dynamic reconstruction of pathway maps and automatically tracks upstream and downstream signaling pathways, significantly improving the depth and interpretability of biological pathway data visualization. During the lecture, attendees also witnessed a live demonstration of how users can interact with the platform by asking questions in natural language, allowing them to quickly pinpoint critical molecular events or functional modules. These iterative enhancements not only solidify Reactome’s position as a high-quality, open-access database but also establish it as a powerful, AI-driven platform supporting cutting-edge systems biology research.

Prof. Wout Bittremieux

University of Antwerp

A living benchmark for advancing AI in proteomics and de novo peptide sequencing

Professor Wout Bittremieux introduced a platform titled "A Living Benchmark to Advance AI in Proteomics De Novo Peptide Sequencing." The platform provides 29 benchmark datasets, comprising a total of 6.4 million annotated spectra, which cover a wide range of species (human, animal, plant, microorganisms), multiple enzymatic digestion methods, and various types of mass spectrometry instruments. Application scenarios include immunopeptidomics, metaproteomics, single-cell proteomics, antibody research, as well as viral post-translational modifications (PTMs).

Ground truth PSMs were obtained using three mainstream proteome database search tools. The platform also integrates the runtime environments and parameter settings of 18 de novo sequencing tools, enabling a systematic comparison of predicted PSMs against true PSMs, as well as the calculation of evaluation metrics at both peptide and amino acid levels. Notably, these de novo prediction tools are continuously updated, ensuring that the platform remains dynamically evolving.

Overall, the platform provides a robust, continuously updated, and standardized benchmark system for AI-driven de novo peptide sequencing, making a significant contribution to advancing de novo sequencing methodologies.

Prof. Wilson Goh Wen Bin

Nanyang Technological University

Harnessing AI and proteomics for mental health diagnostics and prognostics: Toward scalable care in Singapore

Professor Wilson Goh’s presentation introduced a research project leveraging AI and proteomics to enhance mental health diagnostics and prognosis, with the goal of enabling scalable care in Singapore. The report highlighted the global mental health crisis and underscored the urgent need for precise biomarkers and early interventions—particularly in the case of psychiatric disorders.

This study leverages the LYRIKS research cohort to collect comprehensive longitudinal data—encompassing clinical, neuropsychological, and multi-omics information. At its core, the research employs advanced AI algorithms developed specifically for complex data analysis (such as PROJECT, MVIDIA, and OPDEA) to scrutinize proteomic datasets. These methods excel at robustly handling missing data, seamlessly integrating multi-view information, and identifying biologically relevant protein features that can predict conditions like schizophrenia and drug resistance. Ultimately, the goal is to establish a national mental health proteomics platform, combining these cutting-edge tools with electronic health records and digital health data, thereby enabling early screening, risk stratification, and personalized treatment—from community settings all the way to clinical care.

Professor Gao Ge

Peking University

Towards a causality-oriented Cell in silico: From prediction to design

Professor Gao Ge has proposed a vision to build a "silicon-based cellular system oriented toward causality," aiming to move from prediction to design. He pointed out that while multi-omics technologies have unveiled the heterogeneity of cells as functional units and the intricate complexity of gene regulation, deciphering their underlying causal hierarchies remains a significant challenge. His team advocates integrating vast amounts of omics data with cutting-edge AI and machine learning approaches to develop a "causality-driven generative model" grounded in both data-informed insights and knowledge-guided principles. This innovative approach will bridge the gap between raw data and deep biological understanding, ultimately enabling the creation of a "cellular regulatory language model"—a tool capable of simulating and rationally designing cellular regulatory behaviors within a computational framework.

Edited by Allegretti Yuan Hu

Cell Systems

Behind the scenes at Cell Press

This time, presented by Cell Systems Dr. Allegretti Yuan Hu, the scientific editor, delivered a presentation titled "Behind the Scenes at Cell Press," providing a comprehensive overview of Cell Systems and its operational model. As an organization under Elsevier, Cell Systems It boasts an extensive portfolio of journals spanning multiple fields, including life sciences, physical sciences, clinical studies, and environmental science.

The speech detailed the criteria and process for editors evaluating manuscripts, covering aspects such as research scope, significance, and methodological rigor. At the same time, it highlighted several services offered to authors, including pre-submission consultations, manuscript transfer options, and an innovative multi-journal submission system—designed to enhance submission efficiency and provide authors with greater flexibility. Additionally, the report offered practical advice on how to optimize abstracts, figures, and method descriptions, helping researchers streamline their submission process even further. Cell Systems Publishing research findings in these high-impact journals. Cell Systems The journal is dedicated to "rigorously understanding any biological phenomenon" through "quantitative, inference-based research methods" and computational models—aligning perfectly with the cutting-edge research paradigm of building AI-driven cell models, making it an ideal platform for publishing groundbreaking achievements in this field.

Professor Zhou Peixian

Peking University

On the mathematical and algorithmic considerations of AI-driven virtual cell construction

Professor Zhou Peixian's presentation this time primarily explored the future of AI Virtual Cells (AIVC), questioning the current reliance on foundation models inspired by large language models. He argued that the development of AIVC should adhere to three key principles. First, while foundation models based on semantic vectors may seem promising, they are unlikely to provide a complete solution—recent benchmark tests have shown that these models often fail to outperform traditional approaches, particularly in generative tasks such as perturbation prediction. Second, AIVC should integrate biological prior knowledge rather than relying solely on "black-box" models; explicitly modeling processes like the cell cycle, adaptive changes, and growth dynamics can significantly enhance accuracy. The speaker advocated for "biological transport" instead of mere optimal transport, suggesting the use of differential equations combined with neural networks to strike a balance between interpretability and predictive power. Finally, AIVC represents a paradigm shift toward active learning, where models actively guide experimental design rather than passively analyzing data. He also noted that, in scenarios with limited data, diffusion models might outperform language model approaches, and introduced innovative methods like flow matching—techniques that generate models efficiently without the need for explicit simulations. Ultimately, the vision is to create a closed-loop system that seamlessly integrates generative capabilities, mechanistic understanding, and proactive experimental guidance.

Professor Sun Siqi

Fudan University

A controllable foundation model for general and specialized biomolecular structure prediction

Professor Sun Siqi introduced IntelliFold, a controllable foundational model that has achieved breakthroughs in both general-purpose and specialized biomolecular structure prediction. He began by reviewing the advancements in de novo peptide sequencing and protein structure-prediction technologies, highlighting the challenges current models face—particularly in complex scenarios such as antibody-antigen interactions and protein-nucleic acid complexes—when evaluated against the FoldBench benchmark. He then outlined IntelliFold's core strengths: its ability to deliver high-precision predictions across multiple modalities, including proteins and nucleic acids, while maintaining ultra-fast, low-memory computational efficiency. Additionally, IntelliFold can accurately predict allosteric sites and other critical structural features through customizable constraints.

Researcher Wen Han

Beijing Institute for Scientific Intelligence / Peking University

AIVC powered by a multimodal and dynamical foundation model

Professor Wenhan introduced Uni-RNA, the team’s latest large-scale nucleic acid pre-training model. This model has demonstrated outstanding performance in tasks such as RNA structure prediction and functional annotation, significantly enhancing the ability to model omics data at the nucleic acid level. He further explored the potential applications of large-scale foundation models and neural network dynamics in life science omics research, particularly their promising role in complex system modeling and virtual experiment simulations. Building on this, he proposed that AIVC—powered by multimodal and dynamical foundation models—could integrate multi-source heterogeneous data from transcriptomics, proteomics, spatial omics, and microscopy, ultimately enabling the construction of a unified cellular state representation framework equipped with temporal dimensions and context-aware capabilities. He emphasized that establishing AIVC will break through the limitations of traditional static cell modeling, allowing researchers to more accurately simulate cells’ dynamic responses under various stimuli, thereby providing a smarter computational platform for mechanistic exploration, drug screening, and disease modeling.

Dr. Daniel Hornburg

User

Every cell counts, every peptide matters: From large-scale studies to the tiniest single cells, recent advancements in mass spectrometry-based proteomics

Dr. Daniel systematically introduced the technical features of the timsTOF series instruments: They excel in robustness, detection depth, and sensitivity, enabling scalable support for biological research discoveries and facilitating the accumulation of high-quality proteomic data. This technology can precisely distinguish between various types of peptides—such as enzymatically digested peptides, immunopeptides, glycopeptides, activity probe peptides based on ABPP technology, and even subtle chemical background signals—at multiple dimensions. He also provided detailed insights into the applicable scenarios for different timsTOF models: At the single-cell level, the timsUltra AIP technology achieves an impressive detection rate of 500 single cells per second within a 30,000-square-micrometer area, effectively capturing spatially resolved single-cell proteomic data. Meanwhile, the timsOmni platform is specifically designed for analyzing post-translational modifications and protein conformations, significantly enhancing its ability to identify immunopeptides.

Edited by Jiao Yuxia

Genomics, Proteomics & Bioinformatics

Publishing with GPB, a premium journal in the omics field

Genomics, Proteomics & Bioinformatics (Abbreviated as GPB ) is an international journal jointly hosted by the China National Center for Bioinformation and the Genetics Society of China, and published by Oxford University Press. It ranks within the top 10% in the JCR field of Genetics & Genomics, with a recent two-year impact factor of 7.9. In her report, Executive Editor Dr. Jiao Yuxia introduced the journal’s positioning and unique features, emphasizing GPB Dedicated to publishing high-quality, multi-omics research and technological innovations from around the globe, covering a diverse range of content types—including reviews, perspectives, original research, databases, web tools, methods, and protocols. She also highlighted the journal’s author services and academic events, encouraging researchers to actively submit their work and collectively advance the field of omics.