A Most Bitter Enemy
by Mechas
We carry most of our scars deep inside, buried and often difficult to access. These tidbits speak of past encounters and aggressions, contributing to the story of who we are. Just like us, bacteria also bury these pieces of past encounters, but they do so deep in their DNA.
Bacteria are well able to protect themselves from other organisms, a necessary trait if they are to survive amid changing conditions and threats from aggressors. Work in recent years indicates that perhaps the most frequent of these assailants are phages, tiny particles that are nonetheless considered to be the most abundant biological entities on our planet.
Image source, © Graham Beards.
Two recently published studies report advances in the discovery of bacterial immune systems that protect bacteria against viral infection. They do so by developing machine learning algorithms to explore microbial sequence space to predict antiphage defense systems. The premise is that AI tools trained to predict patterns on known datasets can be used to identify antiphage systems, thus expanding the known arsenal of bacterial immune systems.
DeWeirdt et al. from MIT developed a machine learning model, DefensePredictor, capable of accurately identifying antiphage proteins. They trained the model on a set of genes from about 17,000 prokaryotic reference genomes and then tested it on 69 Escherichia coli strains. DefensePredictor identified with high confidence 624 defense-related proteins, 100 of which shared no detectable homology to known defense proteins. Experimental validation of predicted proteins was done by cloning some of these systems in E. coli and testing for protection against a panel of phages. They showed that 42 of the 94 systems tested protected cells against at least one phage. Among these systems, 15 contained novel protein domains not previously shown to be involved in defenses. Predictions done with additional microbial genomes resulted in more than 5,000 predicted defense proteins with no clear homologs to known defense proteins.
Image source.
Mordret et al., from the Institut Pasteur in Paris, took a different approach to discovering novel bacterial immune systems. They developed three complementary deep-learning models to predict antiphage function by leveraging protein information and both sequence and genomic contexts. Experimental validation of predicted systems in E. coli and Streptomyces albus identified 12 antiphage systems that had not previously been linked to antiviral defense. These systems contained both known and novel protein domains not previously associated with antiphage immunity. When applied to more than 32,000 bacterial genomes, these tools predicted 2.39 million antiphage proteins, many of which were previously unrecognized. Based on these results, they believe that an estimated ~1.5% of the genes in a typical bacterial genome are involved in antiviral defense.
AI is a powerful approach for the analysis of large amounts of data. As these studies show, AI tools can uncover previously unrecognized genomic traits; in this case, an abundant and previously unrecognized reservoir of antiviral defense systems. The discovery of novel mechanisms also creates the opportunity for exploring new potential biotechnological applications, as occurred with the gene-editing technology adapted from the CRISPR-Cas system in bacteria and archaea.
It is comforting that science advances despite worldwide efforts to divert our attention or national priorities aimed at diminishing open inquiry and curiosity about our natural world. For this greater understanding, which inspires both awe and creativity, will always be essential for progress and innovation.
We'd be delighted to hear from you! You can comment below, send us an email or reach us via Bluesky or Mastodon. Please subscribe to get email notifications of our new posts.
