Thema der Dissertation:
Exploring feature identification and machine learning in predicting protein-protein interactions of disordered proteins Thema der Disputation:
Machine learning for prediction of paired-data: Pitfalls and strategies
Exploring feature identification and machine learning in predicting protein-protein interactions of disordered proteins Thema der Disputation:
Machine learning for prediction of paired-data: Pitfalls and strategies
Abstract: Machine learning has already been successfully applied to various facets of biological data. However, it is often the case that the employed strategies do not align well with the inherent characteristics of the data. The nature of the datasets used by the machine learning models can introduce additional problems that are not immediately apparent, causing standard evaluation methods to fall short. Understanding and evaluating machine learning models applied to pairs of entities, such as proteins, ligands, or drugs, presents a particularly challenging task due to the nature of the inputs and the lack of consistent and reliable frameworks. In this talk, I will initially discuss about general pitfalls in machine learning models for biological data. Subsequently, I will focus on the flaws in evaluation schemes for pair-prediction models including the challenges such as selection of training and test set, as well as the proposed evaluation schemes for machine learning models.
Zeit & Ort
16.02.2024 | 14:30
Seminarraum 032
(Fachbereich Mathematik und Informatik, Arnimallee 6, 14195 Berlin)