Mgr. Jiří Sedlář, Ph.D.

Mgr. Jiří Sedlář, Ph.D.

Ing. Tomáš Kalvoda, Ph.D.

Department of Applied Mathematics

Stroke is a major cause of death and disability, particularly in low-income countries where access to expensive thrombolytics is limited. Staphylokinase (SAK) could potentially become an affordable and safer alternative to costly treatments; however, its efficacy still needs to be improved. This thesis explores machine learning-guided directed evolution to enhance the activity of SAK. After introducing key biological concepts, the thesis compares current machine learning approaches, selecting EVOLVEpro for its effectiveness in low-data settings. Two datasets, one from Laroche et al.~and one from Loschmidt Laboratories, were analyzed to deepen the understanding of the thrombolytic activity of SAK. The Laroche dataset was also used for training to determine whether it could be utilized for pretraining purposes. However, the model showed limited generalization ability, particularly in predicting high and low activity values. Neither data normalization nor logarithmic transformation improved the performance, most likely due to the imbalanced nature of the Laroche dataset. In parallel, a set of 20 SAK variants, including 15 randomly selected variants and five from the Laroche dataset, all restricted to non-interface mutations, was proposed for testing in the first round of directed evolution at Loschmidt Laboratories.