(in Polish) Analiza danych biomedycznych 1000-5D22ADB

Today's medical challenges often involve diseases with complex genetic and molecular backgrounds. Modern molecular profiling methods yield vast resources of tabular or imaging data. Analysis of these data can help understand how diseases such as cancer or infectious diseases arise, how they work, and how to treat them.

Biomedical data analysis is a very capacious field of research that uses a variety of mathematics and computer science methods: artificial intelligence, machine learning, probabilistic methods, statistics. It is currently a very intensively developing field of interest for both private companies and all leading universities.

The topics of the seminar focus on molecular data analysis methods. Many papers deal with current research projects in which the research groups leading the seminar are involved. Our recent interests include the following topics:
- Antimicrobial resistance. We are developing specialized deep generative models for the generation of synthetic antimicrobial peptides that can kill antibiotic-resistant bacteria. (Methods used: deep learning, generative models.)
- Modeling the tumor microenvironment. What is the spatial organization of the tumor and its neighborhood? How do they interact with each other? These interesting questions can be addressed using spatial transcriptomics, digital tumor imaging or mass spectrometry data. (Methods: probabilistic graphical models, machine learning models.)
- Reconstructing cancer family trees. Which cancer mutations appear first? How do metastases arise? How does drug resistance arise in cancer? Is cancer evolution neutral or driven by selection? These and many other questions about the family history of cancer cells are very exciting for us! (Methods: probabilistic graphical models, mathematical models.)
- Deep Pathologist. Can deep learning algorithms improve the work of pathologists? Artificial intelligence methods, such as convolutional neural networks, can be trained on histological images of tumors to recognize multiple tissue types. (Methods: deep learning models.)
- Modeling drug efficacy. We attempt to understand and predict how drugs act on cancer cell lines. (Methods: statistical models, optimization algorithms.)
- Mutagenic processes in cancer. The mutation landscape of a cancer genome is a result of complex interactions between DNA damage, DNA repair, and other biological processes. Such processes can be studied through the lenses of characteristic mutation patterns imprinted by individual mutagens. We analyze these patterns, link them to specific causes, uncover and model interactions between them. (Methods: statistical data analysis, probabilistic methods, machine learning.)
- Predicting the activity of regulatory sequences. Despite the vast amount of experimental data available on genomic regulatory regions such as enhancers and promoters, predicting the effect caused by even a small mutation is challenging. To better understand the principles of regulatory regions, we are developing convolutional neural networks to predict the activity of any DNA sequence in different cell types. (Methods: statistical data analysis, machine learning.)
- Reproducing gene regulatory networks. The interactions between chromatin, transcription factors and genes form a complex network of interactions that can be reconstructed using single cell sequencing data. (Methods: statistical data analysis, probabilistic methods.)