SciLifeLab AI seminar series
The Scilifelab AI Seminar series aims to facilitate knowledge sharing pertaining to artificial intelligence (AI) among the life science community. The focus is being put on Machine Learning approaches including deep learning neural network models used in data-driven life science projects, but also on integrative data analysis studies, where data from multiple sources are combined to gain novel insights.
The seminar is run over Zoom on the third Friday of the month during academic terms (with some exceptions), typically between 10 and 11 am, with ca. 45 min long presentation and 15 min discussion. Zoom link: http://meet.nbis.se/aiaio.
The seminar series is open to everyone and is jointly organised by the NBIS and SciLifeLab Data Centre.
Speakers from both life science academia and industry are welcome.
To stay updated, you can join our email list by contacting Olga Dethlefsen (olga.dethlefsen@dbb.su.se) or Bengt Sennblad (bengt.sennblad@scilifelab.se).
Current organisers
- Bengt Sennblad (NBIS)
- Olga Dethlefsen (NBIS)
- Erik Ylipää (NBIS)
- Jonas Söderberg (NBIS)
- Mahbub Ul Alam (SciLifeLab Data Centre)
Contact
For questions, contact bengt.sennblad@scilifelab.se
Image by Danielle Navarro, License: CC BY 4.0
Schedule, Autumn 2025
These events can also be found via our Events calendar.
August 22, 16:00-17:00 Note the unusual time!
Vivek Adarsh
Mithrl
Mithrl AI Co-Scientist for Accelerating Biological Discovery
Despite the explosion of biological data from sequencing, imaging, and high-throughput experiments, extracting meaningful insights remains slow, manual, and heavily reliant on limited bioinformatics resources. Scientists often encounter bottlenecks that delay decision-making and limit the scope of discovery.
In this talk, we introduce the concept of an AI Co-Scientist—a set of autonomous agents that collaborates with researchers to analyze data, generate hypotheses, and uncover novel insights in minutes rather than months. We’ll explore real-world use cases where AI systems assist with tasks ranging from experimental planning and data preprocessing to functional analysis and scientific interpretation. We’ll also address practical challenges, including data quality, trust in AI-generated results, hallucinations, and integration into existing research workflows. By the end of the session, you’ll have a clearer understanding of how AI can partner with scientists to accelerate progress in the life sciences.
September 26, 10:00-11:00
Cailean Osborne
Oxford University
Practical tools for facilitating openness in AI research and development: Introducing the Model Openness Framework and the OpenMDW license
In the rapidly evolving field of AI, openness and transparency are essential for promoting collaboration, trust, and reproducibility. In this presentation, Cailean Osborne will introduce two practical tools designed to support openness in AI research and development: the Model Openness Framework (MOF) and the OpenMDW licence. The MOF offers practical guidance for practitioners to identify which components of machine learning models can be shared, and recommends suitable licenses for each. Complementing this, the OpenMDW licence is a novel permissive license tailored specifically for machine learning models, enabling the use, study, modification, and redistribution of open machine learning models and accompanying artifacts, including code, data, and documentation. Together, the MOF and OpenMDW license provide practical tools for advancing openness and transparency in AI, enabling more reproducible, collaborative, and responsible AI research and development.
October 17, 10:00-11:00
Maria Lerm
Linköping University
Training AI models to predict a person’s health status and biological age from DNA methylation data
Precision medicine has long been synonymous with genomics, yet this narrow view overlooks the profound influence of an individual’s lifestyle, environmental exposures, infections, and other stressors on disease risks of common diseases. In fact, the lasting epigenetic marks caused by these stressors predict disease risks far more powerfully than identification of disease-predisposing genetic variants.
In this talk I will tell you about our project in which we train AI models on rich health data and DNA methylome data to make predictions regarding a persons biological age as well as cardiovascular, metabolic and lung health. I will reveal how the unique explainability built into our AI architecture can find causality in cellular processes contributing to ageing on an individual basis.
November 21, 10:00-11:00
Israel Cabeza de Vaca
Uppsala University (Jens Carlsson group)
Rapid traversal of vast chemical space using machine learning-guided docking screens to accelerate drug discovery
The rapid expansion of make-on-demand chemical libraries now offers access to tens of billions of synthetically accessible molecules, creating unprecedented opportunities for structure-based drug discovery. However, screening such ultra-large libraries remains computationally prohibitive, even with state-of-the-art docking methods.
In this presentation, I will describe a hybrid virtual screening strategy that integrates molecular docking with machine learning to efficiently navigate chemical space at the billion-compound scale. The workflow involves docking approximately one million compounds to a target protein and training a classification model to recognize high-scoring molecules. Using conformal prediction, the model guides compound selection from multi-billion-scale libraries, drastically reducing the number of molecules requiring explicit docking.
Among several algorithms evaluated, CatBoost provided the best balance between accuracy and computational efficiency, enabling large-scale applications. When applied to a 3.5-billion-compound library, this approach reduced the computational cost of virtual screening by over three orders of magnitude. Experimental validation identified novel ligands for multiple G protein–coupled receptors, including compounds exhibiting designed multi-target activity.
These results demonstrate how combining AI-based prediction with physics-based docking can make ultra-large-scale virtual screening a practical and powerful tool for modern drug discovery.
December 12, 10:00-11:00
Cancelled
Schedule, spring 2025
These events can also be found via our Events calendar.
January 17, 10:00-11:00
Erik Ylipää
NBIS, Linköping Universtiy
Natural Language Processing for clinical notes
Clinical notes are a cornerstone of healthcare, capturing essential information about symptoms, diagnoses, treatments, and patient outcomes. This rich data source holds immense potential for improving clinical decision-making and patient safety. In this talk, we explore how natural language processing (NLP) can unlock the value of clinical notes for high-impact healthcare applications, including an example from an ongoing project to detect medical implants for MRI safety.
We’ll address the unique challenges of working with sensitive healthcare data, including privacy concerns, data cleaning complexities, and the specialized AI methods required for this domain. Finally, we’ll discuss how federated learning offers a transformative approach to training AI on sensitive data without centralizing it—an essential step toward realizing AI’s potential in healthcare.
February 14, 10:00-11:00
Nikolay Oskolkov
NBIS, Lund University
Is UMAP accurate? Addressing some fair and unfair criticism.
UMAP is a golden standard dimensionality reduction method in single cell biology, yet it has a controversial reputation and is sometimes heavily criticized, see for example [1 – 5]. In particular, the recent Nature publication of All of Us program [6] gave rise to an avalanche of discussions in scientific community regarding the controversial UMAP figure of human populations suggesting that UMAP is not accurate for this purpose. Remarkably, the main criticism of UMAP originates (to the best of my knowledge) from the population genomics community, wile the single cell community seems to be satisfied with the quality of UMAP analysis.
In this talk I will discuss peculiarities of data in single cell and population genomics analyses, and explain some insights from the UMAP algorithm, which could potentially attempt to resolve the contradiction between the two communities and very different research questions studied by the communities. I will also cover the foundations of PCA + tSNE + UMAP algorithms and emphasize their pros and cons for different types of data in Life Sciences.
March 28, 10:00-11:00
Replaced by Automation of research processes with LLMs and AI
April 25, 10:00-11:00
Avlant Nilsson
DDLS fellow, Karolinska Institutet
Towards an interpretable deep learning model of cancer cells.
Cancer emerges from complex molecular interactions that drive pathological cells states. To model these interactions, we develop deep learning frameworks of cellular signaling, gene regulation, and metabolism. Our approach embeds prior knowledge networks into a recurrent architecture, allowing us to capture cellular dynamics by training on omics data across different perturbations and conditions.
At the core of our modeling is a propagator function that predicts the next cell state based on the current state, enabling the simulation of molecular state transitions over time. To ensure biological plausibility, we employ a technic based on feature embedding and superposition that encodes molecular identities in a structured feature space while restricting the predictive input for each molecular state to its direct network connections. These structured embeddings facilitate the use of a universal function allowing generalization across different molecules, cell types, and experimental settings.
By constraining learned representations to known molecular entities, our framework maintains interpretability while enabling data-driven inference. Ultimately, we aim to integrate our models into a unified representation of cellular behavior. Bridging mechanistic understanding with predictive modeling, this work lays the foundation for AI-driven precision medicine, offering new tools to simulate and control cancer cell states.
Download slides (PowerPoint)
Download slides (PDF)
May 23, 10:00-11:00
Phil Ewels
Seqera
Seqera AI: How we’re using LLMs and Agents with Nextflow code.
LLMs are fast becoming an indispensable tool for anyone writing software. But how well do they cope with Nextflow pipelines? Hear how Seqera is building a suite of AI tools that can kickstart Nextflow developer experience and integrate deeply with bioinformaticians’ tool chains. Learn how we’re going beyond simple chat interfaces, with agentic tooling that can fast-track edits without sacrificing accuracy and reproducibility.
History of the seminar series
This seminar series is a merger of two different seminar series that took place at SciLifeLab in past: NBIS AI/IO seminar series as well as SciLifeLab AI Seminar series.
The NBIS AI/IO seminar series started out as the NBIS Integrative Journal Club in 2016. However, the integrative scope was interpreted more and more freely and included other ML approaches such as Neural Nets. In connection to NBIS reorganisation, 2022, which was partly due to the advent of the Data-Driven Life Science (DDLS) initiative but also due to NBIS growth requiring it, the JC became associated with the new NBIS AI and IO tech group. It was time to formalise the extended scope of the journal club and let the name mirror the changes, hence the new name was NBIS Artificial Intelligence and Integrative Omics Seminar Series (AI and IO).
The SciLifeLab AI Seminar Series was organised in 2020 and 2021 by SciLifeLab Data Centre. It combined scientific highlights from SciLifeLab-affiliated researchers and invited experts on the general topic of AI applications in Life Science.