AI improves biology models to better predict cell processes

A new AI system which improves the predictive performance of computational biology models could enable researchers to gain new insight into how cancer cells develop.

The international AdaLab project launched in 2015 with the intention of designing a ‘robot scientist’: an automated lab system which uses artificial intelligence (AI) techniques to conduct and improve cycles of scientific experimentation. 

AdaLab researchers, co-led by Dr Larisa Soldatova from Goldsmiths, University of London, have now used AI to develop a highly advanced computational model showing complex processes of yeast growth.

The yeast S. cerevisiae is studied within biology as a model for human eukaryotic cells (cells with a nucleus enclosed within membranes) and their transformation, in order to understand particular biological processes and concepts in the body. Yeast models are relevant to understanding what happens during cancer development, the immune system, ageing, and other cell developments.

Existing computational models of biological systems are not particularly good at being able to predict what cells might do in future. Constructing a high-fidelity computational model which holds more information and has better predictive ability requires hundreds or thousands of cycles of development, which existing systems biology research cannot do.

Fully automating the process using advanced ideas from AI is necessary to carry out the number of cycles needed to generate a more complex and accurate model, the research team say.

They combined multiple AI tools with integrated laboratory robots and even after just three cycles of model improvement, a prototype model was developed with more complexity and better predictive performance than existing models. 

A report of the research has been published in the journal PNAS and is available online (open access). 

Dr Soldatova said: “The tools used at the moment have no high-level understanding of what they are doing, the same way chess programs don’t know that they are playing chess. One approach to providing this would be to give the system high-level goals to achieve and a high-level planning ability. Another fundamental enhancement would be to give the AI tools the ability to communicate goals and intentions to human scientists.

“We see a future where combinations of software tools, laboratory automation and human scientists will work together to form systems biology models that fully reflect and predict the underlying biology.” 

In June 2019 Dr Soldatova was awarded £476,570 by the EPSRC to lead the ACTION on cancer project at Goldsmiths and create AI which supports clinicians in deciding the most rational personalised cancer treatment. The project is in partnership with the University of Manchester