Designing and Optimizing Deep Learning Methods for Genomic Sequencing Data

MCML Authors

Hüseyin Anil Gündüz

* Former Member

→ Group Bernd Bischl
 Statistical Learning and Data Science

Abstract

This dissertation applies modern deep learning techniques to genomics, introducing new approaches for self-supervised learning, uncertainty quantification, and automated model design. A key focus is the effective use of unlabeled genomic data, highlighted by the development of Self-GenomeNet, a self-supervised method tailored to genomic sequences. The work also presents automated optimization strategies for model architectures and hyperparameters, achieving better results than expert-designed models. Finally, it contributes user-friendly software that supports various genomic data formats and integrates core methods developed in the thesis. (Shortened).

phdthesis

---

Dissertation

LMU München. Apr. 2025

---

Authors

H. A. Gündüz

---

Links

DOI

---

Research Area

 A1 | Statistical Foundations & Explainability