New AI methods for predicting protein properties to improve plant nutrient uptake

Project description

By improving plant nutrient uptake, we can reduce polluting nutrients such as nitrogen in fertilisers, reducing nutrient losses and over-fertilisation. Such research is based on laborious, time-consuming, and costly lab experiments.

The aim of this project is to:

• Develop novel machine learning methods to pre-select potential advantageous molecular target proteins
• Fast-track the hands-on lab work, and
• Increase the probability of engaging in relevant protein regulations.

Our hypothesis is that intrinsic biological property of proteins are encoded in the semantic embeddings that emerge from large-scale deep sequence models applied to protein amino acid sequences. We will exploit these semantic embeddings to predict protein properties using machine learning.

Project Members 

Tommy Löfstedt, Department of Computing Science, Umeå University and Regina Gratz, Department of Forest Ecology and Management, Swedish University of Agricultural Sciences, SLU.