Title:	Using deep learning to analyze complex biological and chemical data
DNr:	Berzelius-2025-300
Project Type:	LiU Berzelius
Principal Investigator:	Erik Kristiansson <erik.kristiansson@chalmers.se>
Affiliation:	Chalmers tekniska högskola
Duration:	2025-10-01 – 2026-04-01
Classification:	10203
Keywords:

Abstract

Artificial intelligence provides new, disruptive means to interpret and analyze complex biological and medical data. In particular, transformers enable analysis of sequential biological data, such as DNA sequences and medical health records, data that has previously been hard to efficiently incorporate in a deep learning perspective. Another example are the graphical neural networks (GNNs), which provide a general framework to describe the often complex dependencies encountered in and between organisms. In this project, we use state-of-the-art AI methodologies to interpret complex biological and medical data. The project combines the development and fine-tuning of dedicated AI methods with the analysis of large volumes of biological data generated by international collaborative experimental groups. This project covers two projects (both funded by DDLS/KAW): a) analysis of the development, spread, and diagnostics of antibiotic-resistant bacteria and b) assessment of the toxicity of chemicals to humans and the environment based on their molecular structure. In the first project, we use graph attention networks (GATs), transformers, and more traditional machine learning to investigate genes that make bacteria resistant to antibiotics. This includes analysis of the DNA sequences, assessment of their spread between bacterial hosts, and development of decision support to guide antibiotic treatment. In the second project, we use fusion transformers to describe chemical structures, genetic differences between organisms, together with other metadata, for accurate prediction of toxicity. Here, our ultimate aim is to replace animal testing with in silico predictions.