A computer program that can predict which genes help bacteria defend against pathogens could lead to the next generation of precision genetic engineering tools.
The artificial intelligence model recognizes the sequence of genes involved in defenses against bacteriophages—viruses that infect bacteria.
These immune systems have been transferred to powerful gene editing technologies, such as CRISPR-Cas, which enable DNA sequences to be precisely cut, altered or deleted within an organism.
The DefensePredictor tool, described in Scienceis available as an open source tool to enable the discovery of many prokaryotic immune systems.
“Identifying new mechanisms of antiphage defense could yield the next generation of precision molecular tools while providing important insights into the ongoing hand-to-hand race between bacteria and phages,” said molecular biologist Michael Laub, PhD, and colleagues at MIT.
The intense pressure to avoid or survive infection has driven the evolution of many antiphage defense systems, including restriction enzymes and CRISPR-Cas systems.
Although antiphage genes are often found in “islands of protection” in prokaryotic genes, this does not always happen and many systems are dispersed or carried by mobile devices such as plasmids, prophages and transposons.
In order to try to model the identification of antiphage proteins, Laub and the team began to look at the genes of about 17,000 prokaryotic organisms.
They mapped the homologues of known defense and nondefense genes and built a picture of the proteins encoded by these genes as well as their four closest neighbors in the genome.
DefensePredictor is trained on this to identify which genes are involved in defense mechanisms.
After working properly in silkit was tested on 69 different Escherichia coli genes and identified 624 different proteins that it confidently predicted were involved in defense, including 154 that shared no homology to known defense proteins.
About half of the identified defense proteins were not encoded in plasmids, prophages, or defense islands, indicating that the model was able to identify multiple genetic mechanisms.
Of the 94 predicted genes tested in the laboratory, 42 provided protection against at least one of the 24 tested, giving a confirmation rate of about 45%.
Fifteen protein regions across these 42 systems had not yet been proven to be conserved, suggesting that new physiological systems remain undiscovered.
Extending DefensePredictor’s predictive capabilities further E. coli of 1000 different prokaryotic genomes revealed more than 5000 predicted defense proteins that were not clear homologs of those already known.
Next Science A research article in the same issue of the journal also showed how AI can discover different types of bacterial defenses.
Ernest Mordret, PhD, from the Pasteur Institute, and colleagues showed how deep learning designs can lead to the discovery of a large antiphage and a large atlas of antiviral bacterial protection.
The team developed three complementary deep learning models to predict antiphage proteins by leveraging genomic context (ALBERTDF), amino acid sequence (ESMDF), or both (GeneCLRDF).
Twelve newly predicted antiphage systems were experimentally validated Escherichia coli and Streptomyces albus.
When applied to more than 30,000 bacterial genes, the models predicted 2.39 million antiphage proteins, of which 85% had no known immune connection, corresponding to approximately 23,000 predicted antiphage families.
All estimates are made freely available through the functional antiphage atlas.
“We have developed deep learning models to predict antiphage systems,” the authors summarize.
“These methods extract information about a protein’s “conservation” from two seemingly orthogonal sources: its genomic context across thousands of genomes, and its amino acid sequence.
“By combining these complementary markers, we move from a fragmented, incomplete view of bacterial defense to a resolved and quantitative understanding of its repertoire.”
#Bacterial #Defense #Expressions #Change #Genetic #Mutation