A team from the University of California, San Diego, trained a model on 500,000 variant sequences to characterize the initiator, the element that triggers the reading of a gene.
For decades, biologists have known that a short DNA motif, called initiator, is the starting point for transcription of a gene. Its exact definition remained unclear: the proposed rules so far did not describe the actual functional sequences. The team of James T. Kadonaga, at the Department of Molecular Biology at the University of California in San Diego, took the problem by data volume rather than theory.
The laboratory synthesized and measured approximately 500,000 sequence variants, then used this experimental game to train a machine learning model to recognize base pair combinations that make an initiator active. Then applied to the human genome, the model regains signature in about 60% of genes — a much higher proportion than previous definitions suggested. The work was conducted with doctoral student Torrey Rhyne-Carrigg.
The impact is double. Clinically, a predictive rule is used to assess whether a mutation in this region will disrupt the activation of a gene, a question that is now difficult to decide for non-coding variants. On the engineering side, the same rule is used to design synthetic promoters whose triggers are finely controlled for biotechnology or gene therapy.
Read the full article on the original source
Go to the source →