A web-based ML pipeline for riboswitch annotation.
Click here for the tool.
RiboSense is an online web-app tool that utilizes an ML pipeline to detect and classify riboswitch sequences in the provided input. The online tool is created using Vue.js and TailwindCSS as main UI framework, and uses opensource libraries to parse sequencing formats like FASTA, GenBank, and SBOL. Tensorflow.js 2.0 is employed for all model predictions in this tool, and it is set up to run locally using your PC's CPU resources.
The pipeline comprises of two models, one called the Detector Model and another called the Classifier Model, and both of these are manually chained as follows
- The given sequence is cropped and padded, and is input to the Detector Model
- The detector model predicts if the input is a riboswitch sequence
- If the detector model predicts that the input is not a riboswitch sequence, the pipeline ends, and the detector's probability for this prediction is returned.
- If the detector model predicts the input to be a riboswitch sequence, the input is then processed and fed to the Classifier Model
- The Classifier Model predicts out of it's 31 trained riboswitch classes for the input
- The classifier output, it's prediction probaility along with the detector's prediction probaility is returned.