An artificial intelligence (AI) model can accurately predict changes in gene expression in any human cell type.
Numerous factors, including disease processes, can influence gene expression within cells. Current methods to identify changes in gene expression rely on retrospective techniques, which analyse past or ongoing cellular processes under controlled conditions. While effective at revealing how cells function and react in specific circumstances, these methods cannot predict future responses. To address this, researchers from Columbia University, New York, developed an AI-driven model with a prospective approach. This enables predictions of gene expression changes across diverse scenarios, focusing on forecasting potential cellular responses.
'Predictive generalisable computational models allow us to uncover biological processes in a fast and accurate way,' said Raul Rabadan, professor of systems biology at Columbia University, New York, and senior author of this study. He continued, 'These methods can effectively conduct large-scale computational experiments, boosting and guiding traditional experimental approaches.'
Existing AI models for gene expression analysis often rely on cancer cell data, limiting their application to normal cellular processes. To overcome this, the researchers developed GET (general expression transformer), an AI model tailored to normal cells. In the study, published in Nature, GET was trained using data from 213 healthy cell types, including adult and fetal cells. After analysing the DNA sequence and structure from this training data, GET accurately predicted gene activity across a variety of cells. This included cells which GET was not trained on, such as astrocytes.
Jian Ma, professor of computational biology at Carnegie Mellon University, Pennsylvania, who was not associated with this study, told The Washington Post that this work 'directly tackles one of biology's major challenges: understanding how the same genome can drive such diverse behaviours in different cell types.'
The team demonstrated GET's applicability across various cell types by applying it to a paediatric leukaemia, B cell precursor acute lymphoblastic leukaemia (B-ALL). Approximately 30 percent of B-ALL cases arise from mutations within the PAX5 gene, although the mechanism behind this was previously unclear. Using GET's predictions, the team identified a key potential interaction between the PAX5 protein and nuclear receptor proteins which drives the disease process in B cells. GET's prediction was subsequently confirmed through experimental validation.
Recently, there have been many advancements in the field of AI for biological analysis (see BioNews 1265, 1261, 1209, 1119, 1105). The team believes that their work will pave the way for future models which employ a predictive approach. Professor Rabadan concluded, 'It's really a new era in biology that is extremely exciting; transforming biology into a predictive science.'
Leave a Reply
You must be logged in to post a comment.