====== 392160 Generative Models in Biomedicine ======

| [[https://ekvv.uni-bielefeld.de/kvv_publ/publ/vd?id=662762345|392160]] | Schönhuth / Pianesi | Summer 2026 | Tue 16-18 in U2-200 (S) and Thu 10-12 in X-E0-220 (Ü) |

==== Content ====

The rapid advancement of Generative Artificial Intelligence (GenAI) has unlocked new possibilities for modelling and understanding complex biomedical data. Generative models, such as Variational Autoencoders (VAEs), Generative Adversarial Networks (GANs), and Diffusion Models, have emerged as powerful tools for tasks such as molecule generation, medical image synthesis, and biological sequence design. \\ 

The seminar will begin with a series of introductory lectures (around 4 or 5) covering the fundamentals of generative modelling and their applications in biomedical contexts. Both foundational and state-of-the-art approaches will be explored, along with their respective use cases and limitations. \\ 

These lectures will be followed by two dedicated sessions on how to write technical reports and how to deliver effective presentations. \\ 

Student seminar presentations will then take place, conducted in small groups of 1-2 students. \\ 

The course will follow a seminar+tutorial format, and as such students will: 
  - Present a chosen paper on a generative model and its biomedical application; 
  - Deliver a final report of around 10 pages; 
  - Weekly submit a short summary (around 500 words) of the presentation held during that week.

The course will be entirely held in English.

==== Contact ====

  * Prof. Dr. Alexander Schönhuth: [[mailto:aschoen@cebitec.uni-bielefeld.de|aschoen@cebitec.uni-bielefeld.de]]
  * Luna Pianesi: [[mailto:lpianesi@cebitec.uni-bielefeld.de|lpianesi@cebitec.uni-bielefeld.de]]

==== Important Links ====

  * Moodle https://moodle.uni-bielefeld.de/course/view.php?id=13394

==== Structure ====

  * Individual presentations (S): 20 minutes, followed by 5 minutes of discussion
  * Report (S): 12 pages long, excluding references (but title, table of content, figures, and tables included)
  * Blog (S): 30-45 minutes of reading time (TBD more precisely in the next few weeks)
  * Coding project (Ü): reproduce the results of the paper you chose to present

==== Papers ====

TBA \\
If there is any problem with accessing a paper, write an email to [[mailto:lpianesi@cebitec.uni-bielefeld.de|Luna Pianesi]].


==== Time table seminar sessions ====

| **Date** | **Topic** |
|14.04.2026 | Introduction to seminar, Course organisation, How to present |
|21.04.2026 | Lecture 1: motivation and the data landscape |
|28.04.2026 | Lecture 2: taxonomy of generative models |
|05.05.2026 | Lecture 3: generative models for sequences |
|12.05.2026 | Lecture 4: generative models for images and clinical text |
|19.05.2026 | Lecture 5: evaluation, limitations, open problems |
|26.05.2026| CANCELLED - How to write reports + Early bird presentation slot (1) |
|02.06.2026| No seminar session |
|09.06.2026| **Presentation**: Rehman Kerimov. "De novo prediction of RNA 3D structures with deep generative models" \\ **Presentation**: Lena Haggenmüller. "SynthSeg: Segmentation of brain MRI scans of any contrast and resolution without retraining" |
|16.06.2026| **Presentation**: Jannis Laurin Müller. "Bio-xLSTM: Generative modelling, representation, and in-context learning of biological and chemical sequences" \\ **Presentation**: Vincent Jian Arvand. "Denoising diffusion probabilistic models for 3D medical image generation" |
|23.06.2026| **Presentation**: Muhammad Ahmed Bashir. "A long-context language model for deciphering and generating bacteriophage genomes" \\ **Presentation**: Breia Ammen. "MolGPT: Molecular Generation Using a Transformer-Decoder Model" \\ **Presentation**: Marvin Wiebe. "Structure-based drug design with equivariant diffusion models" \\ **Presentation**: Mubashir Ali Noor Qazi. "De novo design of protein structure and function with RFdiffusion" |
|30.06.2026| **Presentation**: Matthis Großkreutz. "Generating Multi-label Discrete Patient Records using Generative Adversarial Networks" \\ **Presentation**: Christos Stoupas. "WangLab at MEDIQA-Chat 2023: Clinical Note Generation from Doctor-Patient Conversations using Large Language Models" \\ **Presentation**: Jona Brockhaus. "A vision–language foundation model for the generation of realistic chest X-ray images" \\ **Presentation**: Rodrigo Sebastian Zepeda Lorenzana. "BioBERT: A pre-trained biomedical language representation model for biomedical text mining" |
|07.07.2026| **Presentation**: Kareem Elbanna. "Bilingual language model for protein sequence and structure" \\ **Presentation**: Sarathy Ramanan Ramasamy. "PathLDM: Text conditioned Latent Diffusion Model for Histopathology" \\ **Presentation**: Sishir Rijal. "Deep generative modeling of sample-level heterogeneity in single-cell genomics" \\ **Presentation**: Sri Lakshmi Maddur Satheesh Kumar. "Robust deep learning-based protein sequence design using ProteinMPNN" |