Predicting Gene Activity Changes
A recent study focuses on forecasting changes in gene activity resulting from genetic modifications. When a specific gene is either silenced or activated, it’s generally expected that only the messenger RNA corresponding to that gene will be affected. However, certain genes produce proteins that influence the regulation of multiple other genes, which can lead to adjustments in the activity of numerous genetic entities. Additionally, alterations in gene status can impact cellular metabolism, prompting extensive modifications in overall gene activity.
The complexity escalates significantly when two genes are considered. In many instances, the genes may perform distinct functions, leading to a straightforward additive effect where the changes from each gene’s alteration are summed. Conversely, if there is an interaction between the genes, changes may enhance or suppress one another, yielding unpredictable outcomes.
To analyze these effects, scientists have employed CRISPR gene-editing technology to deliberately manipulate the activity of one or more genes, followed by sequencing all RNA molecules in the cells to observe the resultant changes. This methodology, referred to as Perturb-seq, offers insights into the cellular functions of the modified genes. For researchers Ahlmann-Eltze, Huber, and Anders, this data serves as a foundation for training AI models to anticipate changes in the activity of other genes.
The researchers began by using foundational models and augmented their training with data derived from experiments involving the activation of either one or two genes through CRISPR. This training encompassed data from 100 single gene activations along with 62 cases of dual gene activation. Subsequently, the AI systems were tasked with predicting the outcomes for an additional set of 62 gene pairs. For the sake of comparison, predictions were also generated using two simplistic models: one that assumed no changes would occur and another that predicted an additive effect, positing that the activation of genes A and B would trigger a combination of changes associated with each individual gene.
