MicroRNAs (miRNAs) are small non-coding RNAs that regulate gene expression by binding to complementary sites in messenger RNAs (mRNAs), primarily through pairing between the miRNA seed region (nucleotides 2–7) and the 3' untranslated region (3' UTR) of target mRNAs. While canonical sites are the most effective, non-canonical interactions have been proposed to also mediate repression. However, recent studies show that non-canonical sites do not lead to significant repression, suggesting that most functional miRNA target sites are canonical. To improve the prediction of miRNA target sites, the authors developed a quantitative model using a compendium of experimental datasets, which considers site type and 14 other features. This model outperformed existing models and was as informative as high-throughput in vivo crosslinking approaches. The model is now integrated into the latest version of TargetScan (v7.0), providing a valuable resource for understanding miRNA-regulated gene networks. The study analyzed the efficacy of non-canonical binding sites identified in high-throughput experiments, including CLIP, CLASH, and IMPACT-seq. Despite these sites being enriched for miRNA pairing, they did not lead to significant repression of mRNAs. This suggests that non-canonical sites are not functional in mediating repression. The authors also confirmed that miRNAs can bind to non-canonical sites, but these interactions do not result in detectable repression. This finding supports the focus on canonical sites for predicting miRNA target sites. To improve the accuracy of miRNA target prediction, the authors preprocessed microarray datasets to minimize technical biases and nonspecific effects. They used partial least-squares regression (PLSR) to estimate and remove components of the transcriptome response that were correlated across experiments. This approach reduced the variance in mRNA responses for mRNAs without sites to the transfected miRNA, enhancing the signal for miRNA-mediated repression. The authors also evaluated 26 features, including site conservation, structural accessibility, and mRNA characteristics, to build a regression model for predicting miRNA target sites. They selected 14 robust features through stepwise regression, which were used to develop the context++ model. This model outperformed previous models in predicting miRNA target site efficacy and is now integrated into TargetScan (v7.0), providing a valuable resource for miRNA target prediction.MicroRNAs (miRNAs) are small non-coding RNAs that regulate gene expression by binding to complementary sites in messenger RNAs (mRNAs), primarily through pairing between the miRNA seed region (nucleotides 2–7) and the 3' untranslated region (3' UTR) of target mRNAs. While canonical sites are the most effective, non-canonical interactions have been proposed to also mediate repression. However, recent studies show that non-canonical sites do not lead to significant repression, suggesting that most functional miRNA target sites are canonical. To improve the prediction of miRNA target sites, the authors developed a quantitative model using a compendium of experimental datasets, which considers site type and 14 other features. This model outperformed existing models and was as informative as high-throughput in vivo crosslinking approaches. The model is now integrated into the latest version of TargetScan (v7.0), providing a valuable resource for understanding miRNA-regulated gene networks. The study analyzed the efficacy of non-canonical binding sites identified in high-throughput experiments, including CLIP, CLASH, and IMPACT-seq. Despite these sites being enriched for miRNA pairing, they did not lead to significant repression of mRNAs. This suggests that non-canonical sites are not functional in mediating repression. The authors also confirmed that miRNAs can bind to non-canonical sites, but these interactions do not result in detectable repression. This finding supports the focus on canonical sites for predicting miRNA target sites. To improve the accuracy of miRNA target prediction, the authors preprocessed microarray datasets to minimize technical biases and nonspecific effects. They used partial least-squares regression (PLSR) to estimate and remove components of the transcriptome response that were correlated across experiments. This approach reduced the variance in mRNA responses for mRNAs without sites to the transfected miRNA, enhancing the signal for miRNA-mediated repression. The authors also evaluated 26 features, including site conservation, structural accessibility, and mRNA characteristics, to build a regression model for predicting miRNA target sites. They selected 14 robust features through stepwise regression, which were used to develop the context++ model. This model outperformed previous models in predicting miRNA target site efficacy and is now integrated into TargetScan (v7.0), providing a valuable resource for miRNA target prediction.