Gabriele Schweikert -- Computational Strategies to Analyse, Impute and Interpret Epigenetic Control Mechanisms of Gene Regulation [IN PERSON]
| Dates: | 22 April 2024 |
|---|
| Times: | 14:00 - 15:00 |
|---|
| What is it: | Seminar |
|---|
| Organiser: | Department of Mathematics |
|---|
| Who is it for: | University staff, External researchers, Current University students |
|---|
| Speaker: | Gabriele Schweikert |
|---|
|
Join us for this seminar by Gabriele Schweikert (Dundee) as part of the North West Seminar Series in Mathematical Biology and Data Sciences.
The talk will be hosted in person in room 4.63 of the Simon Building. For those who cannot attend in person the talk will also be streamed via zoom, please contact carl.whitfield@manchester.ac.uk or igor.chernyavsky@manchester.ac.uk for the zoom link, or sign up to the mailing list.
Title: Computational Strategies to Analyse, Impute and Interpret Epigenetic Control Mechanisms of Gene Regulation
Abstract: Epigenomic modifications are reversible chemical marks on top of the DNA, that do not change the underlying sequence itself. Personal, cell-type-specific epigenomes result from a combination of genetic variants and a cellular memory of past cellular events. Epigenetic mechanisms are therefore essential mediators of gene–environment interactions. Functionally, they contribute to the control of current and future transcription and thus play important roles during development, disease progression and ageing. Recently, efforts to record personal epigenomes across tissues have become feasible. However, the large number of assays required for a complete epigenomic map continues to be a limiting factor for personalised epigenomics.
Machine learning approaches are poised to fill this gap. In this talk I will present eDICE, which is based on the transformer architecture and is capable of predicting individual-specific epigenomic landscapes. We achieve high prediction accuracy by learning factorised representations. At the same time, eDICE has unprecedented generalisation capabilities. The complete model fits into GPU memory and does not require complicated training schemes as the number of parameters is several orders of magnitude smaller than in previous models. These are essential preconditions to apply computational imputation for personalised epigenomics and to use these methods at scale.
To subscribe to the mailing list for this event series, please send an e-mail with the phrase “subscribe math-lifesci-seminar” in the message body to listserv@listserv.manchester.ac.uk
Speaker
Gabriele Schweikert
Role: Principal Investigator
Organisation: University of Dundee
Travel and Contact Information
Find event
4.63
Simon Building
Manchester
