netSmooth: Network-smoothing based imputation for single cell RNA-seq

Abstract

Single cell RNA-seq (scRNA-seq) experiments suffer from a range of characteristic technical biases, such as dropouts (zero or near zero counts) and high variance. Current analysis methods rely on imputing missing values by various means of local averaging or regression, often amplifying biases inherent in the data. We present netSmooth, a network-diffusion based method that uses priors for the covariance structure of gene expression profiles on scRNA-seq experiments in order to smooth expression values. We demonstrate that netSmooth improves clustering results of scRNA-seq experiments from distinct cell populations, time-course experiments, and cancer genomics. We provide an R package for our method, available at: https://github.com/BIMSBbioinfo/netSmooth.

Evaluation of colorectal cancer subtypes and cell lines using deep learning

“…Finally, we investigated the usefulness for CRC subtyping of using prior information from protein interaction networks. Several groups, including our own, have previously incorporated gene–gene interactions using a method called network smoothing (26, 27). This is accomplished by allowing binary mutation values to diffuse over a gene network, a process which assigns nonzero “mutation scores” rather than binary mutation values, to genes which either have mutations or interact with mutated genes.…”

“…To carry this out, we used a gene network defining interactions between genes from the STRING db (28), a database of protein–protein interactions. We applied the netSmooth (27) algorithm (see the Materials and Methods section) to the mutation data before passing it to maui and computed Harrell’s c-index, as above. This revealed that network smoothing mutations further improve the clinical relevance of latent factors learned when integrating multiomics data (c = 0.79) (Fig 1G).…”

“…This revealed that network smoothing mutations further improve the clinical relevance of latent factors learned when integrating multiomics data (c = 0.79) (Fig 1G). Other network sources than STRING db, or protein–protein interactomes (PPIs) in general, also result in improvements using netSmooth , whereas networks with similar characteristics but without real information about interactions do not (27).…”

“…We applied netSmooth (27) to the binary mutation matrix before feeding it into the neural network of maui. The method uses the PPI to smooth noisy molecular assays in effect, incorporating prior data from countless previous experiments, to improve the signal-to-noise-ratio.…”