An In-Silico Identification of Anti-CRISPR Proteins by Using Descriptors Derived from the Primary Structures

Genome editing has been revolutionized by the CRISPR-CAS technology. The lack of an off-switch to stop CAS9's activity in the CRISPR-CAS system causes off-target edits and mutations, which in turn cause adversity. Anti-CRISPR (ACR) proteins have been shown to be naturally occurring inhibitors of CAS activity off-switching. The identification of anti-CRISPR proteins has led to the development of additional biotechnological and medicinal techniques. However, ACR's recognition is tough because of its low sequence similarity, lack of homology, and conserved functional domain. In this context, a number of traditional computational and machine-learning techniques based on homology and some restricted features are presently in use. Owing to the unchangeable importance of ACRs, identifying possible ACRs requires a more rigorous and precise methodology than what is currently used, which will raise the number and rate of ACRs that are found. We have used a variety of machine learning models, including Support Vector Machine (SVM), Random Forest (RF), and Extreme Boosting (XGB), to analyze different physio-chemical, compositional, substitutional, and structural characteristics of protein sequences for this specific goal. Upon conducting a comparative performance analysis with existing methods using an external validation dataset and cross-validation over a range of computed metrics, we discovered that SVM with the optimal feature set performed better than the other models and is set up as the central component of ACR-Predictor. Consequently, by increasing the rate of discovery and volume of ACRs with an acceptable margin of success, the study's primary goal was accomplished.