导航分类KEYNOTES
Distinguished Professor Karin Verspoor FTSE FAIDH Executive Dean, School of Computing Technologies， RMIT University | Distinguished Professor Karin Verspoor is Executive Dean of the School of Computing Technologies at RMIT University in Melbourne, Australia. She is a Fellow of the Australian Academy of Technological Sciences and Engineering, a Fellow of the Australasian Institute of Digital Health, and a 2021 “Brilliant Woman in Digital Health”. She was also selected as a finalist in the Women in AI Australia/New Zealand Awards 2022 for “AI in Innovation”. Karin is passionate about using data and AI to improve health outcomes for people. Her work has a specific emphasis on the use of natural language processing to transform unstructured data in biomedicine, ranging from scientific literature to clinical texts, into actionable information. |
Karin held previous posts as Director of Health Technologies and Deputy Head of the School of Computing and Information Systems at the University of Melbourne, as the Scientific Director of Health and Life Sciences at NICTA Victoria Research Laboratory, at the University of Colorado School of Medicine, and at Los Alamos National Laboratory. She also spent 5 years in tech start-ups during the US Tech bubble, where she helped design an early artificial intelligence system. Karin received a BA with a double major in Computer Science and Cognitive Sciences from Rice University in Houston, TX, USA, and completed both a MSc and PhD in Cognitive Science and Natural Language at the University of Edinburgh, UK. Title: AI co-scientists: the evolving role of Artificial Intelligence tools in science Abstract: There have been a number of recent high-profile publications on the use of Artificial Intelligence tools to support scientific discovery ? and in some cases to perform the entire process of science from ideation to data analysis to paper review. Biomedicine is a key area of focus for many of these tools, given the high value of research aimed at improving human health. In this talk I will review some of the history of the use of AI and specifically natural language processing to support science, and take a look at how large language models (LLMs) are now being applied in this context. Along the way I will discuss some of the strengths and limitations of both the old and the new approaches, to inform some thoughts about what the future for AI technologies in science might look like. | |
Professor Kai Ye Xi’an Jiaotong University | Professor Ye Kai is a full-time Professor at the Faculty of Electronic and Information Engineering, Xi'an Jiaotong University, Xi'an, China. Ye held previous posts as an Assistant Professor at The Genome Institute at Washington University in St. Louis, USA, as an Assistant Professor at Leiden University Medical Center in the Netherlands, and as a Postdoctoral Researcher at the European Bioinformatics Institute (EBI) in the United Kingdom under Professor Rolf Apweiler. He earned his PhD (cum laude) from Leiden University in the Netherlands, following his Master's degree from the College of Pharmacy and Bachelor's degree from the College of Life Sciences at Wuhan University. |
He has been invited to participate in multiple international genome projects, including The 1000 Genomes Project and The Cancer Genome Atlas (TCGA). In early 2016, Ye returned to China and established an interdisciplinary research group integrating informatics and life sciences at Xi'an Jiaotong University. Ye's research primarily focuses on the development of computational and artificial intelligence-driven methods for genomic sequence analysis, enabling the detection, characterization, and functional interpretation of structural variations, gene structures, and genome architectures. His work spans two interconnected areas: algorithmic innovation and large-scale genomic data analysis. Research and project experience: Algorithm development - Traditional modeling: Pindel for structural variant detection; the MSIsensor suite (MSIsensor, MSIsensor-pro, MSIsensor-RNA) for microsatellite instability profiling. - Deep learning: SVision, SVision-pro, and Swave for complex and somatic structural variant discovery via the "seq2image" paradigm; ANNEVO for ab initio gene annotation through modeling of distal chromosomal interactions. Large-scale data analysis - Plants: Opium poppy genome project. - Human: Chinese Pangenome Consortium for human population diversity. - Pan-species: Darwin Tree of Life project for broad biodiversity characterization. Seq2image: A Holistic Image-Based Paradigm for Genomic Sequence Analysis Abstract: The analysis of complex biological sequences has historically been dominated by reductionist approaches, wherein elegant statistical models are iteratively patched with numerous ad-hoc filters to handle real-world data complexity, ultimately leading to over-engineered pipelines. To bridge the gap between reductionism and holism, we present seq2image, a conceptual framework that transforms one-dimensional biological sequences into two-dimensional images, thereby leveraging deep learning and advanced image processing techniques to encode both raw data and biological knowledge in a data-driven manner. This talk introduces three major applications of the seq2image paradigm. First, for structural variant (SV) detection, we demonstrate how clustering local alignment features as image patterns enables the resolution of complex SVs (SVision), the discovery of de novo and somatic variants through inter-sample comparison (SVision-pro), and population-scale genotyping of novel and pathogenic alleles from pangenome graphs (Swave). Second, for gene annotation, we introduce ANNEVO, which models distal genomic interactions via Hi-C-inspired two-dimensional maps and integrates both vertical and horizontal evolutionary histories through a Mixture-of-Experts architecture, achieving highly accurate ab initio annotation at speeds around 100x faster than conventional evidence-dependent pipelines. Third, by applying seq2image operators to Hi-C data from 1,025 species, we trace the evolution of three-dimensional genome architectures, revealing that checkerboard compartmentalization correlates with multicellular complexity and that plants and animals have undergone distinct evolutionary trajectories characterized by functional partitioning through architectural transitions. Collectively, seq2image provides a unified, scalable framework for genomic discovery, from local variant classification and distal interaction modeling to the quantification of chromosome-scale genome architecture evolution. | |
