TargetRNA3 uses machine learning to predict targets of small regulatory RNAs (sRNAs) throughout a genome. At its core, TargetRNA3 employs a gradient boosting classification algorithm that has been trained on thousands of evinced interactions between sRNAs and their regulatory targets in various prokaryotes. When making target predictions, the machine learning algorithm uses a variety of features indicative of regulatory interactions, including the thermodynamics of the interaction and potential homologous interactions in related organisms, if available.

As input, TargetRNA3 requires the name of a genome and the sequence of a small regulatory RNA (sRNA). When a user enters a few characters from the genome name, a drop-down box appears from which the user can select their genome of interest. For the sRNA sequence, the sequence may be in FASTA format or not.

By default, TargetRNA3 reports candidate regulatory targets of the sRNA that have a probability greater than or equal to 0.5 and a p-value less than or equal to 0.05. These input parameters can be adjusted. To increase the sensitivity of TargetRNA3 and predict more targets, the probability parameter can be lowered.

As output, TargetRNA3 produces a plot indicating regions of the sRNA that participate in target interactions, a plot indicating regions of the sRNA that are conserved in other genomes, and a ranked list of candidate targets that it predicts for the sRNA. For each candidate target, TargetRNA3 reports:

The main measure of the likelihood that the sRNA interacts with a candidate target is the probability. Probabilities greater than 0.5 indicate that the candidate is more likely than not to be a target, as determined by the machine learning algorithm. Probabilities less than 0.5 indicate that the candidate is more likely not to be a target, as determined by the machine learning algorithm.

Savvy users may note that energies are not perfectly correlated with probabilities. This is because energies are just one of many features used by the machine learning algorithm to determine probabilities. Other features are not reported by the webserver for the sake of concision. Advanced users are encouraged to make use of the source code.

The TargetRNA3 website enables searching any prokaryotic genome listed as reference or representative by NCBI's RefSeq, as well as many others. There are thousands of such genomes. The site does not provide support for searching any/all prokaryotic genomes. There are too many. If you are interested in using TargetRNA3 to search a genome other than those provided, you have two options.

Option #1. Download the source code and run TargetRNA3 on your own machine. The User Manual describes how to add a new genome to TargetRNA3's search.

Option #2. We do our best to support reasonable requests for individual genomes to be added to the TargetRNA3 website. We receive many such requests, so we kindly ask for your understanding and patience . Unfortunately, we cannot process batch requests, e.g., "please add these 20 genomes." If you would like to request that a genome be added to the website, please follow these steps:

TargetRNA3 source code is available on GitHub

TargetRNA3: predicting prokaryotic RNA regulatory targets with machine learning. Tjaden B. Genome Biology, 24:276, 2023.