COPM2023 Conference
On April 26, 2023, the Computational Oncology and Personalized Medicine COPM2023: CROSSING BORDERS, CONNECTING SCIENCE conference was held under the auspices of His Magnificence Rector of the Silesian University of Technology. The event was international and took place remotely. The participants had the pleasure of listening to two special lectures. One of the guests was Dr. Simon Buffler, who discussed the important topic of radiation protection during a speech entitled: Can and should radiological protection be individualised? The second, special lecture was delivered by Dr. Jack Tuszynski, who discussed a very interesting problem of Targeting cancer cells' DNA repair mechanisms: from in silico predictions to in vitro validation.
The conference was also attended by scientists and PhD students from the Department of Engineering and Exploratory Data Analysis. The conference was enriched with speeches by dr inż. Joanna Żyła, mgr inż. Tomasz Strzoda, mgr inż. Katarzyna Sieradzka, mgr inż. Aleksandra Suwalska, and mgr inż. Marek Socha.
Joanna Żyła presented a paper on the integration of classifier results. The presented work contained the results of classification by neural network for radiomic features and logistic regression for metabolomic features for the diagnosis of lung cancer. The results of these classifiers were integrated using statistical integration and the final results outperformed the individual classifier detections.
Tomasz Strzoda gave a speech on replacing the classic bioinformatics solution (mapping) with techniques known from NLP (Natural Language Processing). The whole thing was based on a machine-learning model and RNA sequences. The studies considered the influence of the encoded sequence length on the final efficiency of the model.
Katarzyna Sieradzka discussed the issue of recognizing white blood cell subpopulations for data from single-cell RNA sequencing experiments. The key aspect of the work was the use of feature filtering techniques and the publicly available HDBSCAN clustering tool. On the basis of designated clusters of observations and information on marker genes of white blood cell subtypes, it was possible to identify cell subpopulations very effectively, which was verified using unsupervised learning techniques. The proposed approach enabled the detection of very rare white blood cell subtypes as well.
Aleksandra Suwalska talked about a novel approach for cell-type identification of high-dimensional mass cytometry data. The divisive clustering approach combines PARC with the k-Means algorithm in the expanded feature space created with the Gaussian Mixture Model decomposition of marker expression values. The method was applied to the Tuberculosis dataset that consists of over 10 mln observations resulting in 27 cell subpopulations, including the rare ones.
Marek Socha raised an important topic concerning the analysis of computed tomography images. Computed Tomography images, despite being standardized through Hounsfield units, may differ from each other. This is due to differences in the patient scanning methodology adopted by the institutions and the model of the scanner. The differences become apparent through small shifts in the intensity values of the grey levels, often unnoticeable to the radiologist but having a serious impact on the decision of the AI systems. He proposed a method to normalize voxel intensity values in the CT scan population. The method involves an attempt to linearize individual peaks of voxel intensity values on a histogram against a created reference.
The COPM 2023 conference was not only a very valuable scientific event. Thanks to it, it is possible for many scientists to meet, exchange their views, and share their vast experience in the field of computational oncology and personalized medicine. This allows for the continuous development of scientists and students starting their scientific path.
