Speakers: Paolo Soda, Joana Gonçalves
16 Ottobre 2026 | 10:00
DEIB, Sala Conferenze (Ed. 20A)
Per maggiori informazioni:
Marco Masseroli - marco.masseroli@polimi.it
Silvia Casianelli - silvia.cascianelli@polimi.it
Sommario
As part of the Data Science Seminars – Bioinformatics Focus, two invited talks will take place on Friday, October 16th, 2026 at 10:00 AM, in the Conference Room of DEIB (Building 20A - ground floor) organized by the Data Science for Bioinformatics lab.Paolo Soda, will hold the seminar on: "Learning in an Imperfect World: Multimodal, Generative and Adaptive AI from Medicine to One Health".
Joana Gonçalves, will hold the seminar on:"Machine Learning for Molecular Biomedicine".
"Learning in an Imperfect World: Multimodal, Generative and Adaptive AI from Medicine to One Health" - Paolo Soda
Real-world AI rarely operates under ideal conditions: data are heterogeneous and multimodal, observations may be incomplete, and distributions change across sites and over time. These challenges are particularly evident in healthcare, where images, clinical variables, signals and text provide complementary yet fragmented views of the patient. The seminar will present our research on AI methods designed to integrate heterogeneous information, generate missing observations, and adapt to changing environments. It will discuss multimodal learning, generative models for data augmentation and cross-modal synthesis, virtual imaging and patient digital representations, together with adaptive approaches for distribution shift and deployment-time variability. It will show how these methodological principles apply to medicine and extend beyond towards one health, where learning from sparse, heterogeneous and evolving data is equally central.
"Machine Learning for Molecular Biomedicine" - Joana Gonçalves
Significant advances in molecular biomedicine are transforming how we diagnose, treat, and prevent complex disease. Molecular diagnostics are commonplace. Targeted anti-cancer therapies exploiting molecular vulnerabilities with reduced side effects are clinically successful, and mRNA has replaced viral material in vaccines for COVID-19 prevention. Development of new medicines has historically taken years to decades, and 90% of clinical trials fail for lack of efficacy, but mRNA vaccines have shown how computational models can dramatically speed-up discovery. Achieving similar results for complex diseases such as cancer is ambitious, however crucial to address major burdens to healthy living and patient care, like resistance to treatment. Moreover, we live in exciting times for molecular biology, with multimodal omics atlases and perturbational assays being generated at an unprecedented pace. In this seminar, I will present our research in pattern recognition and machine learning (ML) for turning molecular measurements into mechanistic understanding of biology in health and disease, and into actionable therapeutic strategies for precision medicine, while remaining biologically meaningful, robust, and interpretable. I will discuss how we leverage unique combinations of cellular models, perturbation experiments, omics readouts, and ML algorithms to learn about DNA damage repair (DDR) function, identify tumor-specific DDR deficiencies, and predict promising anti-cancer therapy targets. Methodologically, I will introduce our work on general ML challenges that are particularly relevant in molecular biomedicine, including diversity-guided learning strategies for bias mitigation and explainability for unsupervised latent embeddings.
Biografia
Prof. Paolo Soda graduated cum laude in Biomedical Engineering from the University Campus Bio-Medico (UCBM) of Rome in 2004 and obtained his PhD in Biomedical Engineering (Computer Science area) in 2008 from the same university. He is currently a Full Professor of Information Processing Systems at UCBM and a Visiting Professor in Artificial Intelligence and Biomedical Engineering at the Department of Diagnostics and Intervention, Umeå University, Umeå, Sweden. He also serves as Director of the Italian National PhD Program in Artificial Intelligence, area Health and Life Sciences. He is an expert in artificial intelligence, with research interests including (deep) multimodal learning, generative approaches, explainable artificial intelligence (XAI), and AI algorithm resilience. His research is applied to several domains, including medical imaging, clinical data analysis, P5 medicine, social media analysis, earth observation, and energy management. His scientific activity is documented by more than 230 scientific publications, over 5,350 citations, and an h-index of 39, and an i10-index of 115 (source: Google Scholar, June 2026). He is co-author of eight award-winning papers at international conferences (IEEE LSC 2018, IEEE BIBM 2018, IEEE ICCI*CC 2019, IEEE CBMS 2021, IEEE CBMS 2025, IEEE CBMS 2026). He has been included in the Stanford/Elsevier Top 2% Scientists list for the years 2023, 2024, and 2025 (for annual and/or career-long impact), he was ranked among the top 500 Italian AI experts in 2024, and he is included in the Top Italian Scientists list (topitalianscientists.org). He coordinated the winning teams of two international competitions, the COVID CXR Hackathon (Expo Dubai 2022) and All against COVID-19: Screening X-ray Images for COVID-19 Infection (IEEE 2021). He is a member of IEEE, CVPL, and SIBIM. From 2017 to 2022, he chaired the IEEE International Technical Committee on Computational Life Sciences, and he is a co-founder of two innovative start-ups, BPCOmedia srl and Cognivia srl.Joana Gonçalves is an Assistant Professor in Pattern Recognition & Bioinformatics at Delft University of Technology (TU Delft). She leads a Computational Molecular Biomedicine lab, with research in machine learning to understand biological mechanisms, uncover disruptions driving complex disease, and predict novel therapeutic targets from high-dimensional molecular biology data. Methodologically, the research bridges between algorithmics, statistics, and machine learning, focusing on challenges such as learning robust and interpretable models from scarce, biased, and multimodal data. The lab is one of few European partners in international NIH-funded consortia on multimodal molecular atlases of human tissue (e.g. HuBMAP and KPMP). Joana also (co-)leads national research consortia with the Leiden and Erasmus University Medical Centers on DNA repair and the learning of individual genomic-phenotypic diversity in the population from multi-donor cell villages, funded by Holland PTC and the Convergence Impulse & Flagship initiatives. Joana has supervised 5 PhD candidates (the first 2 graduated in 2025), 4 postdocs, and 30+ Master and Bachelor thesis students to date. Before becoming faculty, Joana was a postdoctoral researcher at the Netherlands Cancer Institute (NKI) and a Marie Curie/ERCIM postdoctoral fellow at the Netherlands Centre for Mathematics and Informatics (CWI). She received a PhD in Computer Science with distinction from the Technical University of Lisbon, funded by a personal grant from Fundação para a Ciência e a Tecnologia, and was a guest researcher during her PhD at the University of Leuven (KU Leuven).