Machine Learning-Driven Discovery of Essential Binding Preference in Anti-CRISPR Proteins

QingLan Ma; YuHang Zhang; Lei Chen; YuShen Bao; Wei Guo; KaiYan Feng; Tao Huang; Yu-Dong Cai

doi:10.1002/prca.70013

Machine Learning-Driven Discovery of Essential Binding Preference in Anti-CRISPR Proteins

Proteomics Clin Appl. 2025 Jun 30:e70013. doi: 10.1002/prca.70013. Online ahead of print.

Authors

QingLan Ma¹, YuHang Zhang², Lei Chen³, YuShen Bao¹, Wei Guo⁴, KaiYan Feng⁵, Tao Huang^{6

7}, Yu-Dong Cai¹

Affiliations

¹ School of Life Sciences, Shanghai University, Shanghai, China.
² Channing Division of Network Medicine, Brigham and Women's Hospital, Harvard Medical School, Boston, Massachusetts, USA.
³ College of Information Engineering, Shanghai Maritime University, Shanghai, China.
⁴ Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences, Shenzhen, China.
⁵ Department of Computer Science, Guangdong AIB Polytechnic College, Guangzhou, China.
⁶ Bio-Med Big Data Center, CAS Key Laboratory of Computational Biology, Shanghai Institute of Nutrition and Health, University of Chinese Academy of Sciences, Chinese Academy of Sciences, Shanghai, China.
⁷ CAS Key Laboratory of Tissue Microenvironment and Tumor, Shanghai Institute of Nutrition and Health, University of Chinese Academy of Sciences, Chinese Academy of Sciences, Shanghai, China.

PMID: 40588792
DOI: 10.1002/prca.70013

Abstract

Purpose: Anti-CRISPR (Acr) proteins can evade CRISPR-Cas immunity, yet their molecular determinants remain poorly understood. This study aimed to uncover key features driving Acr activity, thereby advancing both fundamental knowledge and the rational design of robust CRISPR-based tools.

Experimental design: We compiled a binary-encoded matrix of 761 InterPro-annotated domains and binding-site features for known Acr proteins. Seven feature ranking algorithms were applied to prioritize determinant features, and an incremental feature selection strategy, coupled with four distinct classifiers, was used to identify optimal subsets. Consensus key features were defined by intersecting the top subsets across all methods.

Results: Key identified features include the DUF2829 domain, the Lambda repressor-like domain and Sulfolobus islandicus virus proteins, the Cro/C1-type helix-turn-helix domain, phage protein, and replication initiator A. These findings illuminate novel structural modules and regulatory motifs that underpin Acr inhibition.

Conclusions: This study provides critical theoretical support for deciphering Acr mechanisms and offers actionable insights for engineering next-generation CRISPR-Cas applications in clinical and biotechnological settings.

Summary: The CRISPR system is a part of the antiviral immune defense initially discovered in bacteria and archaea. At present, the CRISPR system has become the cornerstone of genome editing technologies such as CRISPR-Cas9, widely used in clinical, agricultural, and biological research. Anti-CRISPR proteins are a group of proteins that inhibit the normal activity of CRISPR-Cas system in certain bacteria or archaea and avoid having the phages' genomes destroyed by the prokaryotic cells. The anti-CRISPR protein family has various components, but with similar functions to help exogenous DNA escape from the immune system. This study tried to uncover molecular mechanisms for anti-CRISPR proteins.

Keywords: CRISPR; anti‐CRISPR proteins; domain; machine learning.

Abstract

Grants and funding