Springe direkt zu Inhalt

Disputation Jakob Konrad Hertzberg

05.05.2023 | 16:00
Thema der Dissertation:
Identification and Prioritization of Putative Pathogenic Structural Variants based on Functional Annotation
Thema der Disputation:
Transformer Neural Networks and their Application in Bioinformatics
Abstract: The transformer neural network architecture is designed for processing sequential data by utilizing the concept of “self-attention” to focus on relevant parts of the input sequence and computing weighted representations for each element. This allows for capturing long-range dependencies in the data. Originally developed for neural machine translation, transformers have since been applied to a wide range of sequence processing tasks including bioinformatic applications such as predicting protein structures, tissue-specific expression patterns, or quantifying the effects of genomic variants.
In my first talk, I will provide an overview of the machine-learning concepts leading up to the development of transformers, describe the model architecture in detail and discuss recent examples of transformer models for variant prioritization. The second talk will be a summary of my dissertation in which I describe a novel pipeline for the detection and prioritization of potentially disease-causing structural variants in patients with limb malformations.

Zeit & Ort

05.05.2023 | 16:00

Seminarraum 046
(Fachbereich Mathematik und Informatik, Takustr. 9, 14195 Berlin)