OutSingle: a novel method of detecting and injecting outliers in RNA-Seq count data using the optimal hard threshold...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• RNA-sequencing (RNA-Seq) gene expression (GE) outlier detection is crucial for identifying aberrant genes linked to Mendelian disorders.
• Current models often use negative binomial distribution (NBD) but face challenges with computational demands for unbiased parameter inference and confounder control, or use biased methods hindering interpretability.

Purpose of the Study:

• To introduce OutSingle, a computationally efficient and interpretable method for detecting outliers in RNA-Seq GE data.
• To demonstrate OutSingle's superior performance in identifying real and artificial outliers masked by confounders compared to existing methods.

Main Methods:

• OutSingle utilizes a log-normal approach for count modeling and singular value decomposition (SVD) with the optimal hard threshold (OHT) for confounder control.
• The method offers an "almost instantaneous" detection of outliers and enables the injection of artificial outliers for robust testing.

Main Results:

• OutSingle significantly outperforms a state-of-the-art denoising autoencoder model in detecting RNA-Seq GE outliers masked by confounders.
• The method demonstrated superior performance on 16 out of 18 tested datasets, highlighting its robustness and generalizability.
• OutSingle's SVD/OHT approach provides a straightforward and interpretable model.

Conclusions:

• OutSingle offers a computationally efficient, interpretable, and highly effective solution for outlier detection in RNA-Seq GE data.
• The methodology is applicable to broader problems involving outlier detection in matrices with confounding factors.
• The availability of the OutSingle code facilitates its adoption and further research in the field.