AI Technology for Life

In our group we focus on developing AI technology to gain understanding of  biological systems from high throughput measurements.

In order to achieve this, we develop and use a wide range of computational technologies, including:

  • Explainable AI
  • Multi-modal AI
  • Fine-tuning foundation models
  • Bioinformatics workflows
  • Multi-omics integration techniques
  • Simulations of biological systems

We work on various biological problems, including:

  • Unravelling plant-microbiome interactions
  • Medical image analysis
  • Quantification of the human proteome
  • The impact of genomics alterations in cancer
  • Mechanisms of protein aggregation
  • Finding novel biomarkers from multi-omics data

Collaborations

As a group, we are always interested in research collaborations both to develop AI technology and tackle complex research questions on living systems, with research institutes as well as companies. Please contact us if you want to discuss opportunities for collaboration.

Embedding

AI Technology for Life is embedded the Department of Information and Computing sciences (within the AI & Data Science division) and the Department of Biology. There are also strong links with the Department of Pharmaceutical Sciences.

Background

Living systems are very complex, with many levels of regulation. Over the past decades, technology has been developed to measure a huge number of data points. For example, several types of ‘omics’ data can be collected on such systems (e.g. genomics, transcriptomics and proteomics). At the same time, AI methodology has improved tremendously, allowing us to make vastly more accurate predictions for many different tasks.

However, data collected from living systems is different from the types of data for which most current AI algorithms have been developed:

  1. Life science data is more challenging to interpret directly: we do not yet understand the detailed functions and mechanisms of the cell and we do not have an intuitive understanding of what a DNA sequence means.
  2. It is usually  expensive to collect a lot of data (more than 1000 data points) with good annotations for a prediction task of interest (e.g. disease outcome, phenotype, behaviour). AI algorithms require such annotated data in their training procedure.

These obstacles highlight the need for the clever adaptation of AI algorithms, to make them suitable for learning effectively on life science data.

Publications

Please find here a full list of publications of the AI Technology for Life group.