Circ-EnviroPredict: A machine learning-based tool to predict potential involvement of circRNAs with cold and drought stress through a Word2Vec approach

Circular RNAs (circRNAs) are a class of RNAs characterized by a covalently closed loop structure formed between the 5’ and 3’ splice sites. Over the years, circRNAs have been shown to act in post-transcriptional regulation and as potential sponges for miRNAs. Although the number of circRNAs identified has grown and several databases have been developed, challenges persist in analyzing large datasets and extracting molecular information. Machine learning-based tools have been gaining attention due to their potential to massively analyze molecules and understand patterns in response to stresses. However, there is still a lack of studies that connect circRNAs and abiotic stress using this approach. In this study, we developed circ-EnviroPredict, a tool designed to predict the potential involvement of circRNAs in cold and drought stress conditions based on biological sequence data. Using a Random Forest–based methodology, the circ-EnviroPredict tool was trained using rice and maize circRNA sequence data, in which k-mers were transformed into vector representations using the Word2Vec approach. Using independent test sets, it was possible to obtain accuracy values ~77% and ~81% forcold and drought models respectively. It was also possible to validate using data from other plant species, including Arabidopsis thaliana, Glycine max and Triticum aestivum. In addition, a k-mer density analysis revealed an enrichment of AT-rich motifs in circRNAs associated with abiotic stress conditions, providing biological insights into the sequence patterns captured by the predictive models. These results provide insights into how machine learning and Word2Vec techniques can be used to classify the potential involvement of plant circRNAs under abiotic stress conditions using biological sequence data.