GenomeNet - Deep neural nets for genomic modelling¶
In search of the right architecture¶
The joint project GenomeNet is focused on genomics, more specifically on the modelling of genomic data. One of the greatest challenges in the life sciences is to better understand the complexity of the dynamics and organisation of the genome. Despite major progress in recent years, the functional relevance of a large part of the genome is still unknown. A promising approach to gain more insight in the field of genomics is the application of the deep learning method, which has already been employed successfully in other areas. However, different deep learning network architectures are suitable for different types of data representations. So far, it is not yet known which architectures or combinations of them might be suitable exactly for modelling genomes.
This project aims to develop a deep neural network for modelling genomic data. The novel method will use so-called model-based optimisation to find a special deep learning architecture for processing genomes. It is planned to train the network with publicly available genomes. The new network will have a wide range of applications: It will be used for the de-novo identification of still unknown structures, for the imputation (completion) of missing nucleotides as well as for the development of data-efficient classification systems that require less labelled data than before.
This work is funded by the Federal Ministry of Education and Research (Germany).