Background: Intellectual disabilities (IDs) are a group of developmental disorders with high phenotypic and genotypic heterogeneity. Association of genetic elements to IDs has typically been empirically accomplished, however recently, machine learning (ML) has proved to be an excellent instrument to elucidate these associations. miRNAs are short non-coding molecules that participate in spatiotemporal gene regulation, making them relevant for the understanding ID causality. Methods: In this study we used the BrainSpan spatio-temporal expression database to develop a series of machine learning predictors: SVM, RF, FF-ANN, and Stochastic Gradient Descent Classifier. These models were capable of recognizing gene expression profiles. The best classifier was used to label miRNAs associated with NS-IDs using the BrainSpan expression profiles. Results: The model with the best performance was a FF-ANN with 0.78 of F1-score, 0.78 of weighted recall and 0.78 of weighted precision. We used this model to identify miRNAs with high probability to be associated with NS-IDs using the spatio-temporal gene expression profile in the human brain. Labeled miRNAs that were annotated were associated with processes related to either IDs and-or neurodevelopmental processes. Conclusions: The development of a machine learning framework that identified potential NS-ID miRNAs represents an interesting approach for the identification of a potential list of on genes that could be subject for further experimental validation. This study also reinforces the potential of machine learning frameworks in their discovery of potential biomarkers that could improve disease detection and management. Keywords: miRNA association; artificial intelligence; machine learning; intellectual disability; biomarker

Background: Intellectual disabilities (IDs) are a group of developmental disorders with high phenotypic and genotypic heterogeneity. Association of genetic elements to IDs has typically been empirically accomplished, however recently, machine learning (ML) has proved to be an excellent instrument to elucidate these associations. miRNAs are short non-coding molecules that participate in spatiotemporal gene regulation, making them relevant for the understanding ID causality. Methods: In this study we used the BrainSpan spatio-temporal expression database to develop a series of machine learning predictors: SVM, RF, FF-ANN, and Stochastic Gradient Descent Classifier. These models were capable of recognizing gene expression profiles. The best classifier was used to label miRNAs associated with NS-IDs using the BrainSpan expression profiles. Results: The model with the best performance was a FF-ANN with 0.78 of F1-score, 0.78 of weighted recall and 0.78 of weighted precision. We used this model to identify miRNAs with high probability to be associated with NS-IDs using the spatio-temporal gene expression profile in the human brain. Labeled miRNAs that were annotated were associated with processes related to either IDs and-or neurodevelopmental processes. Conclusions: The development of a machine learning framework that identified potential NS-ID miRNAs represents an interesting approach for the identification of a potential list of on genes that could be subject for further experimental validation. This study also reinforces the potential of machine learning frameworks in their discovery of potential biomarkers that could improve disease detection and management.