Bacteriophage Genetic Edition Using LSTM

Area of Science:

• Computational Biology and Bioinformatics
• Synthetic Biology
• Antimicrobial Research

Background:

• Bacteriophages (phages) are potent antimicrobial agents against multi-drug resistant bacteria.
• Natural phages have limitations including narrow host specificity, structural fragility, and immunogenicity.
• Systematic methods for genetically engineering phages to overcome these limitations are lacking.

Purpose of the Study:

• To explore the application of artificial intelligence (AI) for guiding and accelerating phage genome modification.
• To develop AI-driven tools for generating phage variants with improved characteristics, such as expanded host range.
• To address challenges hindering the prophylactic and therapeutic use of bacteriophages.

Main Methods:

• Proposed an AI architecture with two deep learning components: a phage-bacterium interaction predictor and a phage genome-sequence generator.
• The predictor utilized a 1-D convolutional neural network (1D-CNN) to analyze phage and bacterial genomes.
• The generator employed a long short-term memory (LSTM) recurrent neural network to modify phage genomes for host range improvement.

Main Results:

• The AI generators achieved an average training accuracy of 96.1%.
• Modified genomes improved the host range for an average of 18 out of 42 phages studied.
• Average host range increased by 73.0% and 103.7%, with maximum host ranges extending from 21 to 24 and 29.

Conclusions:

• Deep learning methodologies can effectively guide the genetic modification of bacteriophages.
• AI approaches demonstrate significant potential for engineering phages with enhanced properties, such as extended host range.
• This work validates the use of AI in overcoming limitations of natural phages for therapeutic applications.