From Code to Cure: Building Computer-Aided Phage Therapy Pipelines

When antibiotics fail, doctors are running out of options. Phage therapy offers an alternative: viruses that can be calibrated to kill specific bacteria.

I recently read the article, “Advances in the field of phage-based therapy with special emphasis on computational resources,” by Gajendra Raghava. It gave me insights into the actual process required to bring computational phage therapy to life, along with the different streams in which phage therapy is being advanced through technology. To define phage therapy simply, it is the use of bacteriophages, viruses that kill bacteria, to target and precisely take down harmful bacteria. Despite the strength that this technology presents, a key issue that still exists is: how can we efficiently identify, evaluate, and approve effective phages for treatment? This is where technology plays a role, integrating computational tools, machine learning, and bioinformatics to accelerate the process of transforming phage data into treatments.

I decided to encapsulate the pipeline through a detailed review of three papers that each highlight different approaches to the computational phage therapy pipeline. Together, they illustrate the current progression of phage therapy and future steps embedded with challenges to overcome.

The first article, “Advances in the field of phage-based therapy with special emphasis on computer-aided phage therapy,” by Gajendra Raghava, lays out the foundation for computer-aided phage therapy(CAPT). It depicts the steps essential for a modern pipeline to cover:

1. Starting with high-quality databases of phage and bacterial genomes
2. Next, the developments of computational prediction models of phage-host interactions
3. Then, the screening of genomes for safety issues, such as virulence or lysogenic genes
4. Reporting results for further experimental validation

This review encompasses a large issue currently: while many individual tools exist, a big challenge with them is integration into a seamless workflow. So, the main goal of CAPT is to create systems that flow well together to guide real-world therapeutic decisions.

The second article, “Sphae: an automated toolkit for predicting phage therapy candidates from sequencing data” by Bhavya Papudeshi et al., demonstrates a working CAPT pipeline. Sphae is designed to analyze raw sequencing data of phages, assemble genomes, and then rapidly screen them for therapeutic potential. It flags unsafe candidates by detecting antimicrobial resistance genes, virulence factors, or lysogeny markers, while filtering out low-quality assemblies or mixed genomes, and it does all this in just 10 minutes. It shows the great potential of CAPT in clinical settings as a powerful tool for quick target identification.

The third article, “An ensemble method for prediction of phage-based therapy against bacterial infections” by Suchet Aggarwal et. al., focuses on one of CAPT’s most difficult steps: phage-host matching. The authors introduce PhageTB, an ensemble system that combines multiple approaches: BLAST, CRISPR spacer matching, and machine learning models based on k-mer frequencies. By combining these methods, the model achieves higher accuracy than any single tool. While PhageTB doesn’t address safety screening directly, it solves one of the critical early hurdles in phage therapy pipelines: knowing which phages can infect which bacteria.

Looking across these three studies, the pieces of CAPT begin to fit together. PhageTB has host prediction; Sphae ensures safety, genome quality, and usability; The Briefings in Bioinformatics review shows how these steps can and should be integrated. Challenges will remain, but this progress is promising and gives hope for the eventual development of a pipeline able to give personalized cures for any bacterial strain within minutes.

Sources: https://academic.oup.com/bib/article/24/1/bbac574/6961791 https://academic.oup.com/bioinformaticsadvances/article/5/1/vbaf004/7959522 https://www.frontiersin.org/journals/microbiology/articles/10.3389/fmicb.2023.1148579/full