New AI improves predictions of guide RNA efficiency in CRISPRi

Many CRISPR technologies are used specifically to alter or silence genes and inhibit protein production. One of these tools is CRISPRi (CRISPR interference), which blocks genes and gene expression without changing the DNA sequence. Similar to the traditional CRISPR mechanism, the guide RNA directs a nuclease (Cas). However, CRISPRi binds to the nuclease DNA without cutting it, resulting in the reduction of the corresponding gene. CRISPRi has become a leading technique to silence gene expression in bacteria. However, design rules are poorly defined.

Until now, predicting the efficacy of CRISPRi for a specific gene has been challenging. But researchers have now developed a machine learning approach using data assimilation and artificial intelligence (AI) to improve such predictions in the future. The scientists used data from multiple genome-wide CRISPRi essentialization screens to train the machine learning approach. Their goal was to better predict the efficiency of engineered guide RNAs used in the CRISPRi system.

The authors found that gene-specific properties of targeted genes had a significant impact on guide RNA depletion in genome-wide screens. In addition, combining data from multiple CRISPRi screens significantly improves the accuracy of prediction models and enables reliable estimates of guide RNA efficiency. By predicting guide RNA efficiency, this study provides valuable insights for designing more effective CRISPRi experiments by implementing precise gene-silencing strategies.

This work has been published Genetic BiologyIn the paper, „Improved prediction of bacterial CRISPRi guide efficiency from depletion screens by mixed-effects machine learning and data assimilation.„

„Unfortunately, genome-wide screens provide only indirect information about guide efficiency. Therefore, we used a new machine learning method that extracts the efficiency of the guide RNA from the impact of the silenced gene,” explains Lars Barquist, PhD, Research Group Leader at the Würzburg Helmholtz Institute for RNA-Based Infection Research (HIRI). and Junior Professor at the Faculty of Medicine at the University of Würzburg.

The team „developed a mixed-effects random forest regression model that provides better estimates of guide performance.” In doing so, they established understandable design rules for future CRISPRi experiments. The study authors validated their approach by conducting an independent screen targeting essential bacterial genes, showing that their predictions were more accurate than previous methods.

„The results show that our model outperforms existing methods and provides more reliable predictions of CRISPRi efficacy when targeting specific genes,” said Yanying Yu, a PhD student in Barquist's research group.

The scientists were particularly surprised to find that the guide RNA itself was not the primary factor determining CRISPRi depletion in essential screens. „Some gene-specific traits related to gene expression are more influential than previously thought,” explained Yu.

The study also reveals that combining data from multiple data sets significantly improves prediction accuracy and enables more reliable assessment of the performance of guide RNAs.

„Expanding our training data by combining multiple experiments is necessary to develop better prediction models. Prior to our study, lack of data was a major limiting factor for prediction accuracy,” Barquist noted. „Our study provides a blueprint for developing more precise tools to manipulate bacterial gene expression and ultimately help us better understand and fight pathogens.”