CRISPR–Cas9 gRNA efficiency prediction: an overview of predictive tools and the role of deep learning

Abstract

The clustered regularly interspaced short palindromic repeat (CRISPR)/CRISPR-associated protein 9 (Cas9) system has become a successful and promising technology for gene-editing. To facilitate its effective application, various computational tools have been developed. These tools can assist researchers in the guide RNA (gRNA) design process by predicting cleavage efficiency and specificity and excluding undesirable targets. However, while many tools are available, assessment of their application scenarios and performance benchmarks are limited. Moreover, new deep learning tools have been explored lately for gRNA efficiency prediction, but have not been systematically evaluated. Here, we discuss the approaches that pertain to the on-target activity problem, focusing mainly on the features and computational methods they utilize. Furthermore, we evaluate these tools on independent datasets and give some suggestions for their usage. We conclude with some challenges and perspectives about future directions for CRISPR–Cas9 guide design.

CRISPRedict: The case for simple and interpretable efficiency prediction for CRISPR-Cas9 gene editing

“…Therefore, we created two versions of CRISPRedict to reflect the different experimental conditions in the guide design process. The training datasets were chosen based on their size, homogeneity, and previous evaluation results [18, 21, 28]. Similarly, to compare the performance of CRISPRedict with the state-of-the-art methods on datasets of different expression systems, we collected 12 datasets from independent studies.…”

“…First, we defined the initial feature set using sequence characteristics that have been shown to influence cleavage efficiency [15, 18, 19, 28–32, 34, 35]. This includes overall and position-specific nucleotide composition, as well as features that reflect the structural properties of the guide sequence.…”

“…For learning-based tools, in particular, these features are combined by models that are generated through machine learning [12, 17]. Thus, they seem to perform better than alignment-based and hypothesis-driven tools [18]. Moreover, a shift has been observed recently in the learning-based category.…”

“…However, the accuracy of machine learning-based tools varies widely among the constructed features and the testing datasets [18, 24]. Moreover, the hand-crafted features may result in redundancy, leading to inaccurate and non-interpretable predictions.…”

“…Furthermore, complex models are not necessarily better than simpler ones. Previous studies have, in fact, shown that the right method may differ per task and conventional machine learning tools can outperform the deep learning ones [18, 26]. Therefore, choosing the proper algorithm depends on the task and may not be based solely on predictive accuracy.…”