A Transformer-Based Model to Predict Micro RNA Interactions

The prediction of the interactions of micro RNA (miRNA) for the regulation of the cellular biological processes represents a challenging bioinformatics problem, with important implications for the design of new RNA-based drugs. We present miRInter-Trans, a model that integrates the RNA-FM foundation model pre-trained on a large corpus of non coding RNA (ncRNA) data with a feed-forward neural network trained on the RNA-FM hidden representations of ncRNA sequences. The model is able to successfully predict miRNA interactions using only the sequence of the ncRNA pairs. To our knowledge, this is the first work addressing ncRNA–ncRNA interaction prediction using sequence alone and embeddings from an RNA foundational model. The proposed approach demonstrates superior performance compared to a state-of-art Minimum Free Energy method.