Prof. Yasset Perez‑Riverol
EMBL’s European Bioinformatics Institute
Quantms ecosystem: data, formats and algorithms to generate AI‑ready proteomics datasets
Professor Yasset Perez‑Riverol systematically outlined the construction goals and key technical approaches of the Quantms ecosystem. He pointed out that current proteomics data still exhibit significant disparities in scale, format, and quality control, which greatly limit their usability in artificial intelligence models. To address this, Quantms employs standardized data structures, scalable algorithmic modules, and automated processing workflows to efficiently convert raw mass spectrometry data into “AI‑ready” structured inputs. Professor Perez‑Riverol elaborated on the strategic optimizations in the ecosystem’s core components, including data cleaning, retention time calibration, feature extraction, and quantitative consistency control, while emphasizing the critical role of format unification in enhancing data reusability and cross‑project interoperability. In addition, Quantms supports seamless integration with various downstream machine learning frameworks, providing a highly flexible and reproducible analytical platform for proteomics research. This presentation offered important technological pathways and practical references for the deep integration of AI and proteomics.
—
Prof. Chris Sander
Harvard Medical School
Machine learning for proteomic perturbation biology
Professor Chris Sander articulated the vision of “perturbation biology”: to build computable and predictable virtual cell models through systematic perturbation of cells and measurement of multidimensional responses. He noted that the challenge lies in moving beyond single‑protein measurements, integrating temporal data and domain knowledge, and redefining cellular processes in a data‑driven manner. His team employs AI methods—such as protein structure prediction and design models like EVcouplings—to guide enzyme engineering and drug design (e.g., target flexibility, combination therapies), and extends these approaches to disease prevention, for instance using electronic health records to predict pancreatic cancer risk. The ultimate goal is to construct multi‑scale executable models ranging from cells to tissues and organs, enabling precision medicine that moves from prediction to intervention.
—
Dr. Arunima Singh, Editor
Nature Methods
Publishing in Nature Methods and pursuing an editorial career
Dr. Arunima Singh, Senior Editor at Nature Methods, delivered an insightful presentation that detailed the journal’s primary focus areas, editorial team composition, and article types. She further shared the key elements editors consider during manuscript evaluation, including topic, novelty and significance, practical value and generality, validation, and application, and systematically explained the writing techniques for Methods articles. Dr. Singh emphasized that Nature Methods’ core focus is on the method itself—authors should “make the method the star.” She advised researchers to place their work in a broader disciplinary context, not only outlining the state‑of‑the‑art but also highlighting how their method goes beyond existing technologies, while providing direct comparisons and demonstrating concrete applications. She also reminded that reproducibility of methods and accessibility to editors and reviewers are equally crucial. Moreover, Dr. Singh introduced the daily work of a professional editor and the experience and skills required to become a scientific journal editor, such as improving writing and reading abilities, staying informed about broad research trends, and networking with editors. Her sharing provided valuable guidance and inspiration for researchers aspiring to submit to Nature Methods or pursue careers in scientific publishing.
—
Prof. Henning Hermjakob
European Bioinformatics Institute, EMBL‑EBI
Reactome 4: pathways reimagined – dynamic visualisation and intelligent chat
Image
Professor Henning Hermjakob introduced the major updates in version 4 of the Reactome database, highlighting the newly launched dynamic pathway visualization and the integrated intelligent chat assistant, designed to enhance user interaction and analytical efficiency. The new Reactome supports dynamic reconfiguration of pathway diagrams and automatic tracing of upstream and downstream signal propagation paths, significantly enriching the depth and interpretability of biological pathway data visualization. Live demonstrations showed how users can interact with the platform through natural language queries to rapidly locate key molecular events or functional modules. These iterative upgrades not only further consolidate Reactome’s status as a high‑quality open database but also make it a powerful intelligent platform supporting systems biology research.
—
Prof. Wout Bittremieux
University of Antwerp
A living benchmark to advance AI in proteomics de novo peptide sequencing
Image
Professor Wout Bittremieux presented a platform entitled “A Living Benchmark to Advance AI in Proteomics De Novo Peptide Sequencing.” This platform provides 29 benchmark datasets containing a total of 6.4 million annotated spectra, covering multiple species (human, animal, plant, microorganism), various enzymatic digestion methods, and different types of mass spectrometry instruments. Application scenarios include immunopeptidomics, metaproteomics, single‑cell proteomics, antibody research, and viral post‑translational modifications (PTMs). Ground‑truth PSMs were obtained through three mainstream proteomics search engines. The platform simultaneously integrates runtime environments and parameter settings for 18 de novo sequencing tools, enabling systematic comparison of predicted PSMs against true PSMs and calculation of evaluation metrics at both peptide and amino acid levels. Notably, these de novo prediction tools are continuously updated, keeping the platform dynamically evolving. Overall, this platform provides a powerful, continuously updated standardized benchmarking system for AI‑based de novo peptide sequencing and represents 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: Towards scalable care in Singapore
Image
Professor Wilson Goh’s presentation introduced a research project leveraging AI and proteomics to improve mental health diagnostics and prognosis, aiming to achieve scalable care in Singapore. He highlighted the global mental health crisis and the urgent need for precision biological diagnostics and early intervention, especially for psychosis. The study is based on the LYRIKS research cohort, which collects complex longitudinal data (clinical, neuropsychological, multi‑omics). The core approach involves developing sophisticated data‑centric AI algorithms (such as PROJECT, MVIDIA, OPDEA) to analyze proteomic data. These methods robustly handle missing data, integrate multi‑view information, and identify biologically relevant protein signatures that can predict conditions such as schizophrenia and drug resistance. The ultimate goal is to establish a national mental health proteomics platform that combines these tools with electronic health records and digital health data, thereby enabling early screening, risk stratification, and personalized treatment from community to clinical settings.
—
Prof. Ge Gao
Peking University
Towards a causality‑oriented Cell in silico: From prediction to design
Image
Professor Ge Gao presented a vision of constructing a “causality‑oriented in silico cell,” aiming to move from prediction to design. He noted that although multi‑omics technologies have revealed cellular heterogeneity and the complexity of gene regulation, challenges remain in deciphering the underlying causal hierarchical relationships. His team advocates combining massive omics data with cutting‑edge AI/machine learning methods to build a “causality‑oriented generative model” based on data‑driven and knowledge‑guided principles. This approach would bridge the gap from data to cognition, ultimately realizing a “cell regulatory language model” that can simulate and rationally design cellular regulatory behaviors in silico.
—
Dr. Allegretti Yuan Hu, Editor
Cell Systems
Behind the scenes at Cell Press
Image
This presentation, titled “Behind the Scenes at Cell Press,” was delivered by Dr. Allegretti Yuan Hu, Scientific Editor at Cell Systems, and provided a comprehensive overview of Cell Systems and its operational model. As an Elsevier publication, Cell Systems is part of a broad portfolio of journals covering life, physical, clinical, and environmental sciences. The talk detailed the editorial criteria and processes for manuscript evaluation, including scope, significance, and methodological rigor. It also highlighted several author services, such as pre‑submission inquiries, manuscript transfers, and an innovative multi‑journal submission system designed to improve submission efficiency and offer authors more choices. Additionally, practical advice was provided on optimizing abstracts, figures, and method descriptions to facilitate publication in high‑impact journals like Cell Systems. The journal’s commitment to “rigorously understanding any biological phenomenon through quantitative, inference‑based approaches” and computational models aligns closely with the research paradigm of building AI‑driven cell models, making Cell Systems an ideal venue for publishing work in this cutting‑edge field.
—
Prof. Peijie Zhou
Peking University
On the mathematical and algorithmic considerations of AI virtual cell construction
Image
Professor Peijie Zhou’s presentation addressed the future of AI Virtual Cells (AIVC), questioning the current reliance on foundation models inspired by large language models. He argued that AIVC development should follow three key principles. First, foundation models using semantic vectors may not be a complete solution, as recent benchmarks indicate they often fail to outperform traditional methods, especially in generative tasks such as perturbation prediction. Second, AIVCs should incorporate biological priors rather than functioning as pure black boxes—explicitly modeling processes such as cell cycle, adaptive changes, and growth dynamics can improve accuracy. He advocated for “biological transport” instead of simple optimal transport, using differential equations combined with neural networks to balance interpretability and predictive power. Finally, AIVCs represent a paradigm shift toward active learning, where models guide experimental design rather than passively analyzing data. He also suggested that diffusion models may outperform language model approaches in data‑limited scenarios, and introduced simulation‑free methods like flow matching for efficient generative modeling. The ultimate vision is to create closed‑loop systems that combine generative capabilities, mechanistic understanding, and active experimental guidance.
—
Prof. Siyi Sun
Fudan University
A controllable foundation model for general and specialized biomolecular structure prediction
Image
Professor Siyi Sun introduced IntelliFold, a controllable foundation model that achieves breakthroughs in both general and specialized biomolecular structure prediction. He first reviewed the development of peptide de novo sequencing and protein structure prediction, and noted the challenges current models face in complex scenarios such as antibody‑antigen and protein‑nucleic acid complexes, as evaluated under the FoldBench benchmark. He then highlighted the core advantages of the IntelliFold model: support for high‑precision prediction across multiple modalities including proteins and nucleic acids, ultra‑fast and memory‑efficient computation, and the ability to accurately predict critical structures such as allosteric sites through custom constraints.
—
Dr. Han Wen
Beijing Institute of Scientific Intelligence / Peking University
AIVC enabled by multimodal and dynamical foundation model
Image
Dr. Han Wen introduced the team’s newly developed large‑scale nucleic acid pre‑training model, Uni‑RNA. This model demonstrates outstanding performance in RNA structure prediction and functional annotation, significantly enhancing the modeling capacity for nucleic acid‑level omics data. He further explored the potential applications of large‑scale foundation models and neural network dynamics in life science omics research, particularly in complex system modeling and virtual experimental simulations. Building on this, he proposed that AIVC, enabled by multimodal and dynamical foundation models, could integrate heterogeneous data sources including transcriptomics, proteomics, spatial omics, and microscopy imaging to construct a unified cellular state representation framework with temporal and contextual awareness. He emphasized that the establishment of AIVC would overcome the limitations of static cell modeling, enabling researchers to more realistically simulate cellular responses under various stimuli and providing a more intelligent computational platform for mechanistic exploration, drug screening, and disease modeling.
—
Dr. Daniel Hornburg
Bruker
Every cell counts, every peptide matters: From large scale studies to smallest single cells, recent advancements in mass spectrometry‑based proteomics
Image
Dr. Daniel Hornburg systematically presented the technical characteristics of the timsTOF series instruments, highlighting their outstanding robustness, detection depth, and sensitivity, which enable large‑scale biological discovery and the accumulation of high‑quality proteomic data. This technology accurately discriminates among different peptide types across multiple dimensions, including enzymatic peptides, immunopeptides, glycopeptides, ABPP‑based activity‑based probe peptides, and chemical background signals. He also detailed the application scenarios for different timsTOF models: at the single‑cell level, the timsUltra AIP technology effectively acquires spatial single‑cell proteomics data by achieving detection rates of 500 single cells per second within a 30,000 µm² region; while the TimsOmni platform focuses on post‑translational modifications and proteoform analysis, significantly enhancing immunopeptide discovery capabilities.
—
Dr. Yuxia Jiao, Editor
Genomics, Proteomics & Bioinformatics
Publishing with GPB, a premium journal in the omics field
