Unit:
Department of Informatics and TelecommunicationsΠληροφορική
Author:
KALIMERIS ALEXANDROS
Supervisors info:
Ιωάννης Εμίρης, Καθηγητής, Τμήμα Πληροφορικής και Τηλεπικοινωνιών, Εθνικό Καποδιστριακό Πανεπιστήμιο Αθηνών
Original Title:
Deep Learning on Point Clouds for 3D Protein Classification Based on Secondary Structure
Translated title:
Deep Learning on Point Clouds for 3D Protein Classification Based on Secondary Structure
Summary:
Proteins are macromolecules that regulate a vast amount of biological processes. The spatial structure of proteins is the main determinant of their biological function. Consequently, discovering and researching new efficient methods for classifying proteins based on their 3D shape is an important task with applications in many scientific fields. In the context of this thesis, we explore and examine the capabilities of Deep Neural Networks in performing classification tasks on the complex 3D shapes of proteins. For these purposes, we analyze existing deep learning architectures that showed promising results. Additionally, we test the effectiveness of these architectures by performing a series of protein classification experiments. In our experiments, we represent the geometric 3D shape of proteins as point clouds, a flexible geometric data representation. Also since proteins have different sizes and the deep learning architectures we explore do not consume dynamic size input, we test ways of normalizing the proteins into the same constant size. Finally, we comprehensively present and evaluate the results of our work.
Main subject category:
Technology - Computer science
Keywords:
Protein Classification, Deep Neural Networks, Point Clouds, Secondary Protein Structure, 3D geometrical shape recognition
