Sparsely-connected autoencoder (SCA) for single cell RNAseq data mining

Abstract

Single-cell RNA sequencing (scRNAseq) is an essential tool to investigate cellular heterogeneity. Thus, it would be of great interest being able to disclose biological information belonging to cell subpopulations, which can be defined by clustering analysis of scRNAseq data. In this manuscript, we report a tool that we developed for the functional mining of single cell clusters based on Sparsely-Connected Autoencoder (SCA). This tool allows uncovering hidden features associated with scRNAseq data. We implemented two new metrics, QCC (Quality Control of Cluster) and QCM (Quality Control of Model), which allow quantifying the ability of SCA to reconstruct valuable cell clusters and to evaluate the quality of the neural network achievements, respectively. Our data indicate that SCA encoded space, derived by different experimentally validated data (TF targets, miRNA targets, Kinase targets, and cancer-related immune signatures), can be used to grasp single cell cluster-specific functional features. In our implementation, SCA efficacy comes from its ability to reconstruct only specific clusters, thus indicating only those clusters where the SCA encoding space is a key element for cells aggregation. SCA analysis is implemented as module in rCASC framework and it is supported by a GUI to simplify it usage for biologists and medical personnel.

In silico drug repurposing in COVID-19: A network-based analysis

“…For each disease, we extracted the largest connected component from the subnetwork composed of genes that were modulated in that specific disease condition and verified whether these genes presented a statistically significant ability to generate a disease module. For subsequent analyses, we selected only the diseases that satisfied this module hypothesis, in accordance with the organizing principles of network medicine [3] , [9] , [39] . Notably, we found that for all diseases analyzed (with the exception of ankylosing spondylitis and chronic spontaneous urticaria), the deregulated genes formed statistically significant modules ( Table 1 ).…”

“…Even though the cause-effect relationship cannot be directly inferred by expression data, it is reasonable to assume that disease co-modulated genes are functionally coordinated in response to an external stimulus, implying that they might be part of the same pathways or biological functions, and may influence each other or be influenced by the same underlying mechanism(s). Inspired by the organizing principles of the network medicine paradigm, we evaluated whether disease deregulated genes had the propensity to aggregate in local disease-specific neighborhoods of the human interactome, thus being functionally-related genes displaying a statistically significant tendency to form dense disease modules [3] , [9] , [39] . The accurate identification and localization of these disease modules represents the first step toward a systematic understanding of molecular-level pathological mechanisms, together with the prediction of novel disease-disease relationships.…”

“…In order to test whether this subnetwork forms a statistically significant disease module, for each analyzed disease we randomly selected groups of proteins in the human interactome with the same size and degree distribution as the original list of disease deregulated genes and the following three metrics were computed: 1) the size of the largest connected component (LCC); 2) the number of interactions in the LCC; and 3) the total number of interactions. The three metrics were then z-score normalized by applying a degree-preserving randomization procedure, expecting a p value ≤ 0.05 for genes forming a statistically significant disease module [39] . Log 2 FC thresholds were chosen to guarantee the topological organization of disease deregulated genes in statistically significant modules.…”

A multi-approach and multi-scale platform to model CD4+ T cells responding to infections

“…159 microarray datasets were integrated with 20 proteomic datasets to define the gene activity profiles (matrices) of the five phenotypes. The profiles and the generic human metabolic model Recon 2.2.05 [ 23 ] were used to build the constraint-based models by the GIMME method in the COBRA toolbox [ 48 ]. In summary, the Th0 model contains 4234 metabolic fluxes and 2452 metabolites; Th1, 4160 and 2377; Th2, 4674 and 2623; Th17, 5223, and 2833; and Treg, 3854 and 2178.…”

A Survey of Autoencoder Algorithms to Pave the Diagnosis of Rare Diseases

“…Another example of the use of VAE in single cell data is scVAE [ 11 ] and SCA [ 66 ], which are employed for classification/clustering tasks. scVAE uses different types of VAE with either a Gaussian or a Gaussian-mixture latent variable prior.…”