Title:	Deep learning models for modelling genetic variation
DNr:	Berzelius-2025-301
Project Type:	LiU Berzelius
Principal Investigator:	Carl Nettelblad <carl.nettelblad@it.uu.se>
Affiliation:	Uppsala universitet
Duration:	2025-10-01 – 2026-04-01
Classification:	40402
Homepage:	https://github.com/kausmees/GenoCAE
Keywords:

Abstract

We are developing deep learning models based on autoencoder architectures for modelling genetic variation, as well as predicting traits of economic importance in plant and animal breeding applications. We have recently established our contrastive methods to be competitive in an accepted article in Genetics. Another new development we're considering is to use deep learning embedding techniques for genome assembly tasks. Initial attempts are highly promising and almost nothing has been published along these lines before. During the fall of 2025, we're mainly interested in continuing along the lines of full genome embeddings (the original scope), k-mer assembly models (as submitted to NeurIPS) and the new promising line of imputation using diffusion models ("genome inpainting").