Cell Subtype Classification via Representation Learning Based on a Denoising Autoencoder for Single-Cell RNA Sequencing

Abstract

Identification of single-cell subtypes is one of the fundamental processes required to understand a heterogeneous population composed of multiple cells, based on single-cell RNA sequencing data. Previously, cell subtype identification was mainly carried out by dimension reduction and clustering approaches that grouped cells with similar expressed profiles together. However, for high robustness to noises and systematic annotation of the subtype in each cell, supervised classification approaches have been widely used. Recently, deep neural network (DNN) models have been widely presented in various fields, including biology. By capturing the composite relationship between sample features and target outcomes, a DNN model enables significant performance improvements in biological data mining analyses. In this paper, we constructed a DNN model, called scDAE for single-cell subtype identification combined with representative feature extraction using a multilayer denoising autoencoder (DAE). The feature sets were learned by the DAE and were further tuned by fully connected layers using a softmax classifier. The model was compared against four state-of-the-art cell subtype identification methods and two conventional machine learning algorithms. From multiple tests, scDAE significantly outperformed competing methods especially on data sets having a large number of cell subtypes and noises. Extracted cell features from the proposed model were clearly clustered with respect to subtype. The results of the experiments indicated that our proposed model is effective in identifying single-cell subtypes and molecular signatures representative of each cell subtype. scDAE is publicly available at https://github.com/cbi-bioinfo/scDAE.

Heuristic Analysis of Genomic Sequence Processing Models for High Efficiency Prediction: A Statistical Perspective

“…A similar representation for Covid and cancer genome analysis can be observed (Fig. 5 ), wherein BiLSTM CNN [ 63 ] and EdeepVPP [ 55 ] have exhibited good accuracy for Covid genomic dataset, whereas VFM [ 56 ], CNV Bayesian [ 46 ], GeneXNet [ 38 ], DAE [ 48 ], SVM [ 48 ], and Active NN [ 48 ] have been shown to outperform others for cancer genome data analysis.…”

“…Similar models have been discussed, wherein copy number variation (CNV) detection, tumour classification using AI, and cell subtype classification using denoising autoencoder (DAE) have been performed on multiple gene datasets [ 46 ]. The CNV approach uses Bayesian inference models for obtaining an accuracy of 99.27% [ 47 ], while AI classifies kidney renal clear cell carcinoma (KIRC), lung adenocarcinoma (LUAD) [ 48 ], lung squamous cell carcinoma (LUSC), and uterine corpus endometrial carcinoma (UCEC) with 96.9% accuracy, which makes it useful for clinical applications [ 49 ]. The artificial intelligent model uses a combination of binary particle swarm optimization and decision tree (PSODT) with CNN, which assists in optimum feature selection, thereby reducing computational delay [ 47 ].…”

“…In Table 2 , it can be observed that EdeepVPP [ 55 ], GeneXNet [ 38 ], RF [ 15 ], VFM [ 56 ], DNN [ 31 ], ResNet [ 38 ], GA SIWR [ 2 ], GBM [ 15 ], DenseNet [ 38 ], PLS [ 15 ], and BiLSTM CNN [ 63 ] outperform others in terms of precision performance, while EdeepVPP [ 55 ], GeneXNet [ 38 ], BiLSTM CNN [ 63 ], RIPPER SVM [ 54 ], RF [ 15 ], VFM [ 56 ], DNN [ 31 ], CNV Bayesian [ 46 ], ResNet [ 38 ], DAE [ 48 ], GA SIWR [ 2 ], GBM [ 15 ], and SVM [ 48 ] have better recall performance than others. While, in terms of computational complexity, GA SIWR [ 2 ], FP Tree [ 2 ], Random Selection [ 60 ], RF [ 15 ], VFM [ 56 ], SVM [ 48 ], PLS [ 15 ], PSODT [ 47 ], and Spectrometry [ 60 ] outperform others; thus, they are categorized as high-speed genome processing algorithms. Fig.…”

Efficient Deep-Learning-Based Autoencoder Denoising Approach for Medical Image Diagnosis

“…where x i is an input value. During the training phase, the proposed denoising autoencoder model was trained to reduce the reconstructive error and increase the chances of recovering inputs [34], since both the encoder and decoder are non-linear. It was trained by minimizing loss function through backpropagation to select the strongest features.…”