Beyond Sequence: A Novel Image-Based Model for MicroRNA Target Prediction

MicroRNAs are small non-coding RNAs that regulate gene expressions in plants and animals. They bind to a gene and stop the process of protein synthesis from the targeted gene. They, through this process, control cell activities such as proliferation, differentiation, and death rate. To understand the functionality of a microRNA, we need to know the targets of the microRNA. Often biologists do in vivo experiments to identify microRNA's targets. However, these experiments are expensive, time-consuming, and limited. In the last two decades, bioinformaticians have developed countless computational models that predict a microRNA's target based on sequence information. But these models mainly suffer from high false positive rates and mediocre recall rates, due to incomplete understanding of the mechanism of microRNA targeting. In this paper, we propose the Base Pair Probability Image (BPPI) model; a novel approach that converts a microRNA target prediction question into an image detection problem. We create an image for each pair of microRNA and target (or non-target) sequences, where the pixel values represent the probabilities of interactions between nucleotides from the two sequences. Then, we use a deep learning model to detect images depicting a microRNA and its target. Our model outperforms several widely used microRNA target prediction software tools, such as miRanda and RNAhybrid. This work has the potential to be used by biologists to shortlist potential microRNA targets for further verification through in vivo experiments.