scKDGM: KAN-guided Dynamic Graph Masked Learning for Single-Cell RNA-seq Clustering

# scKDGM: KAN-guided Dynamic Graph Masked Learning for Single-Cell RNA-seq Clustering
Source: [https://arxiv.org/html/2606.28459](https://arxiv.org/html/2606.28459)
###### Abstract

Single\-cell RNA sequencing \(scRNA\-seq\) clustering is essential for identifying cell types, but high dimensionality, sparsity, dropout, and technical noise hinder robust expression representation and cell graph construction\. Existing masked autoencoders mainly use expression recovery for feature reconstruction, while graph clustering methods usually depend on fixed KNN graphs and do not feed recovered expression back into graph optimization\. We propose scKDGM, a KAN\-guided dynamic graph masked learning framework for scRNA\-seq clustering\. scKDGM uses graph\-aware distribution preserving gene masking \(GDP\-Mask\) to perturb cell identity, a KAN\-based TAKGCN encoder to learn masked\-view representations, mask\-guided expression recovery to construct a dynamic graph, and cross\-view contrastive learning to transfer recovery signals into topology updates\. A ZINB loss models overdispersion and zero inflation\. Experiments on 12 real scRNA\-seq datasets show that scKDGM outperforms 10 baselines in average NMI and ARI\.

## IIntroduction

Single\-cell RNA sequencing \(scRNA\-seq\) profiles transcriptomic heterogeneity at single\-cell resolution and supports cell type identification, trajectory analysis, and disease microenvironment studies\. Early single\-cell whole\-transcriptome sequencing\[[48](https://arxiv.org/html/2606.28459#bib.bib1)\]and high\-throughput platforms such as Drop\-seq\[[40](https://arxiv.org/html/2606.28459#bib.bib2)\]and 10x Genomics\[[97](https://arxiv.org/html/2606.28459#bib.bib3)\]made large\-scale scRNA\-seq routine\. As data scale and tissue complexity grow, unsupervised cell subpopulation identification remains central to scRNA\-seq analysis\[[37](https://arxiv.org/html/2606.28459#bib.bib4),[22](https://arxiv.org/html/2606.28459#bib.bib5)\]\. Related structured\-recognition studies in biomedical and visual domains also show the value of attention, tensor decomposition, fuzzy clustering, and multiscale features under noisy observations\[[17](https://arxiv.org/html/2606.28459#bib.bib81),[96](https://arxiv.org/html/2606.28459#bib.bib89),[80](https://arxiv.org/html/2606.28459#bib.bib73),[9](https://arxiv.org/html/2606.28459#bib.bib78)\]\.

scRNA\-seq data are high\-dimensional, sparse, and dropout\-prone, so direct clustering on Euclidean distances or low\-dimensional projections is sensitive to noise\. Seurat\[[47](https://arxiv.org/html/2606.28459#bib.bib6)\], SC3\[[23](https://arxiv.org/html/2606.28459#bib.bib7)\], and CIDR\[[34](https://arxiv.org/html/2606.28459#bib.bib8)\]improve stability with nearest\-neighbor graphs, consensus clustering, or implicit imputation, but still rely on handcrafted features or shallow assumptions\. Deep methods learn richer representations through embedded clustering, ZINB or negative\-binomial autoencoding, metric learning, masked estimation, and contrastive learning\[[53](https://arxiv.org/html/2606.28459#bib.bib9),[7](https://arxiv.org/html/2606.28459#bib.bib10),[83](https://arxiv.org/html/2606.28459#bib.bib11),[54](https://arxiv.org/html/2606.28459#bib.bib14),[55](https://arxiv.org/html/2606.28459#bib.bib13),[11](https://arxiv.org/html/2606.28459#bib.bib12)\]\. Graph methods model cell relations with GNNs, graph autoencoders, adaptive graphs, and graph contrastive objectives\[[56](https://arxiv.org/html/2606.28459#bib.bib15),[12](https://arxiv.org/html/2606.28459#bib.bib16),[84](https://arxiv.org/html/2606.28459#bib.bib17),[85](https://arxiv.org/html/2606.28459#bib.bib18),[74](https://arxiv.org/html/2606.28459#bib.bib19),[61](https://arxiv.org/html/2606.28459#bib.bib20),[25](https://arxiv.org/html/2606.28459#bib.bib21),[52](https://arxiv.org/html/2606.28459#bib.bib22)\]\. Biological\-prior graphs in scPriorGraph\[[6](https://arxiv.org/html/2606.28459#bib.bib38)\], pathway\-consensus graphs in scMCGraph\[[21](https://arxiv.org/html/2606.28459#bib.bib37)\], and masked graph autoencoding\[[19](https://arxiv.org/html/2606.28459#bib.bib23)\]further improve representation learning\. Broader graph studies similarly emphasize spatiotemporal message passing, high\-order filters, global dependency learning, node collaboration, graph pooling, attributed clustering, community search, graph transitions, metapath associations, and graph\-regularized factorization\[[59](https://arxiv.org/html/2606.28459#bib.bib41),[58](https://arxiv.org/html/2606.28459#bib.bib42),[14](https://arxiv.org/html/2606.28459#bib.bib43),[95](https://arxiv.org/html/2606.28459#bib.bib45),[4](https://arxiv.org/html/2606.28459#bib.bib80),[81](https://arxiv.org/html/2606.28459#bib.bib72),[33](https://arxiv.org/html/2606.28459#bib.bib64),[13](https://arxiv.org/html/2606.28459#bib.bib79),[73](https://arxiv.org/html/2606.28459#bib.bib75),[35](https://arxiv.org/html/2606.28459#bib.bib95)\]\. Yet most graph clustering pipelines still use a fixed KNN graph, which can form hubs, weaken rare or low\-degree cells, and turn expression recovery into reconstruction\-only supervision\.

The expression recovery branch is also related to high\-dimensional incomplete \(HDI\) representation learning\. Reviews and adaptive\-divergence latent factor models show that objective design matters under sparse observations\[[20](https://arxiv.org/html/2606.28459#bib.bib67),[91](https://arxiv.org/html/2606.28459#bib.bib49),[93](https://arxiv.org/html/2606.28459#bib.bib51)\]\. PI/PID control, fuzzy SGD, asynchronous or accelerated parallel SGD, ADMM, Nesterov acceleration, PSO, genetic search, and coevolutionary optimization provide related strategies for stable latent\-factor learning\[[26](https://arxiv.org/html/2606.28459#bib.bib40),[88](https://arxiv.org/html/2606.28459#bib.bib46),[27](https://arxiv.org/html/2606.28459#bib.bib44),[87](https://arxiv.org/html/2606.28459#bib.bib48),[45](https://arxiv.org/html/2606.28459#bib.bib94),[44](https://arxiv.org/html/2606.28459#bib.bib91),[43](https://arxiv.org/html/2606.28459#bib.bib97),[98](https://arxiv.org/html/2606.28459#bib.bib88),[29](https://arxiv.org/html/2606.28459#bib.bib92),[39](https://arxiv.org/html/2606.28459#bib.bib58),[38](https://arxiv.org/html/2606.28459#bib.bib63),[63](https://arxiv.org/html/2606.28459#bib.bib98)\]\. Nonlinear HDI models based on randomized or nonnegative factors, hash factors, multimetric features, autoencoders, outlier\-resilient reconstruction, and prediction sampling further motivate robust feature recovery from noisy high\-dimensional matrices\[[86](https://arxiv.org/html/2606.28459#bib.bib52),[90](https://arxiv.org/html/2606.28459#bib.bib53),[66](https://arxiv.org/html/2606.28459#bib.bib55),[70](https://arxiv.org/html/2606.28459#bib.bib90),[2](https://arxiv.org/html/2606.28459#bib.bib93),[68](https://arxiv.org/html/2606.28459#bib.bib65),[65](https://arxiv.org/html/2606.28459#bib.bib82),[69](https://arxiv.org/html/2606.28459#bib.bib96)\]\.

The dynamic graph component is motivated by temporal and tensor representation studies\. Kalman filtering, temporal QoS modeling, neighborhood regularization, temporal bias, and traffic imputation show how sparse observations can be refined by temporal structure\[[94](https://arxiv.org/html/2606.28459#bib.bib39),[89](https://arxiv.org/html/2606.28459#bib.bib50),[92](https://arxiv.org/html/2606.28459#bib.bib54),[78](https://arxiv.org/html/2606.28459#bib.bib68),[76](https://arxiv.org/html/2606.28459#bib.bib74),[79](https://arxiv.org/html/2606.28459#bib.bib86)\]\. Mode\-aware Tucker networks, neural Tucker factorization, dynamic graph mixers, attention\-based or neural tensor factorization, and battery\-life tensor prediction indicate that nonlinear and mode\-aware tensor models can capture interactions beyond static features\[[72](https://arxiv.org/html/2606.28459#bib.bib57),[49](https://arxiv.org/html/2606.28459#bib.bib62),[18](https://arxiv.org/html/2606.28459#bib.bib76),[50](https://arxiv.org/html/2606.28459#bib.bib85),[3](https://arxiv.org/html/2606.28459#bib.bib60),[77](https://arxiv.org/html/2606.28459#bib.bib71),[30](https://arxiv.org/html/2606.28459#bib.bib77),[8](https://arxiv.org/html/2606.28459#bib.bib83)\]\. Dynamic transaction networks, tensor causal convolution, convolution\-bias factorization, fine\-grained tensor regularization, electricity\-theft detection, and spatiotemporal recovery further support linking recovered signals with evolving relational structure\[[31](https://arxiv.org/html/2606.28459#bib.bib66),[32](https://arxiv.org/html/2606.28459#bib.bib69),[60](https://arxiv.org/html/2606.28459#bib.bib70),[71](https://arxiv.org/html/2606.28459#bib.bib87),[42](https://arxiv.org/html/2606.28459#bib.bib56),[67](https://arxiv.org/html/2606.28459#bib.bib84)\]\.

We propose scKDGM, a KAN\-guided dynamic graph masked learning framework for robust scRNA\-seq clustering\. scKDGM uses a KAN\-based Topology Adaptive Graph Convolutional Network \(TAKGCN\) encoder\. KAN replaces fixed activations and scalar linear weights with learnable univariate edge functions\[[36](https://arxiv.org/html/2606.28459#bib.bib25)\], and KAN\-based graph networks extend this idea to graph learning\[[5](https://arxiv.org/html/2606.28459#bib.bib26),[28](https://arxiv.org/html/2606.28459#bib.bib27)\]\. TAKGCN also follows TAGCN\-style multi\-hop aggregation\[[10](https://arxiv.org/html/2606.28459#bib.bib24)\]; modular graph convolution, graph tensor attention, and network compression studies provide related evidence that graph operators and compact neural transformations can improve structured representation learning\[[15](https://arxiv.org/html/2606.28459#bib.bib61),[57](https://arxiv.org/html/2606.28459#bib.bib47),[16](https://arxiv.org/html/2606.28459#bib.bib59)\]\. On this basis, GDP\-Mask samples non\-neighbor donor cells and shuffles values within each gene column\. The model recovers expression from the masked view, builds a differentiable dynamic graph from the recovered matrix, and aligns masked and dynamic views by contrastive learning\. Thus expression recovery directly refines graph topology instead of acting only as reconstruction supervision\. Our contributions are:

- •First, we propose TAKGCN, which aggregates high\-order cell\-neighborhood information through topology\-adaptive graph convolution and models nonlinear gene\-expression dependencies with Fourier KAN\-style transformations\.
- •Second, we design GDP\-Mask to construct graph\-aware and distribution preserving cell identity perturbations for masked recovery\.
- •Third, we introduce a mask\-recovery\-driven dynamic graph learning mechanism that uses recovered expression information to refine cell adjacency\.
- •Experiments on 12 real scRNA\-seq datasets show that scKDGM achieves the best average NMI and ARI under the benchmark setting\.

## IIMethods

### II\-AOverview

Given an scRNA\-seq matrixX∈ℝn×gX\\in\\mathbb\{R\}^\{n\\times g\}, scKDGM learnsZ∈ℝn×dZ\\in\\mathbb\{R\}^\{n\\times d\}and partitionsnncells intoCClatent groups\. Preprocessing removes zero\-count genes and cells, applies library\-size normalization, median scaling, log transformation, and Scanpy\-based selection of the top 1000 highly variable genes\[[64](https://arxiv.org/html/2606.28459#bib.bib36)\]\. A KNN graphA0A\_\{0\}initializes message passing\.

Fig\.[1](https://arxiv.org/html/2606.28459#S2.F1)shows the two\-stage pipeline\. During masked pre\-training, GDP\-Mask generatesX′X^\{\\prime\}and an observed\-change maskMMfromXXand the current graphAtA\_\{t\}\. TAKGCN encodesX′X^\{\\prime\}onAtA\_\{t\}, predictsMM, reconstructsX^\\hat\{X\}, and usesX^\\hat\{X\}to build a differentiable dynamic graph\. Clean expression is re\-encoded on this graph and aligned with the masked view\. During clustering refinement, the mask branch is removed, DEC\-style clustering is optimized on clean expression, and dynamic\-graph contrastive learning remains active\.

![Refer to caption](https://arxiv.org/html/2606.28459v1/x1.png)Figure 1:The architecture of scKDGM\. GDP\-Mask produces the graph\-aware masked feature matrix and observed\-change mask\. TAKGCN encodes the masked expression on the current graph, while the recovered feature matrix is used to construct dynamic graph\.
### II\-BGDP\-Mask

GDP\-Mask perturbs expression by graph\-aware, gene\-wise shuffling\. For cellii, non\-neighbor donors underAtA\_\{t\}form

𝒟t​\(i\)=\{r∣At​\(i,r\)=0,r≠i\}\.\\mathcal\{D\}\_\{t\}\(i\)=\\\{r\\mid A\_\{t\}\(i,r\)=0,\\ r\\neq i\\\}\.\(1\)If𝒟t​\(i\)\\mathcal\{D\}\_\{t\}\(i\)is empty, the donor is sampled from non\-self cells\. Withπt​\(i\)∼Uniform​\(𝒟t​\(i\)\)\\pi\_\{t\}\(i\)\\sim\\mathrm\{Uniform\}\(\\mathcal\{D\}\_\{t\}\(i\)\), GDP\-Mask samples

Si​j∼Bernoulli​\(ρ\),S\_\{ij\}\\sim\\mathrm\{Bernoulli\}\(\\rho\),\(2\)whereρ\\rhois the mask rate, and replaces only within the same gene column:

Xi​j′=\{Xπt​\(i\)​j,Si​j=1Xi​j,Si​j=0\.X^\{\\prime\}\_\{ij\}=\\begin\{cases\}X\_\{\\pi\_\{t\}\(i\)j\},&S\_\{ij\}=1\\\\ X\_\{ij\},&S\_\{ij\}=0\\end\{cases\}\.\(3\)This preserves each gene’s marginal distribution while introducing non\-neighbor cell identity confusion\. The supervised target is the observed\-change mask

Mi​j=𝕀​\(Xi​j′≠Xi​j\)\.M\_\{ij\}=\\mathbb\{I\}\(X^\{\\prime\}\_\{ij\}\\neq X\_\{ij\}\)\.\(4\)Thus supervision ignores unchanged sampled entries, such as0→00\\rightarrow 0, and forces recovery of cell\-specific expression from graph context and gene combinations\.

### II\-CTAKGCN Encoder

TAKGCN combines TAG\-style multi\-hop aggregation with KAN\-style nonlinear transformations\. LetH\(ℓ\)H^\{\(\\ell\)\}be the layer input andA~\\tilde\{A\}the normalized adjacency\. One TAKGConv layer is

H\(ℓ\+1\)=∑r=0KΦr​\(A~r​H\(ℓ\)\)\+b\.H^\{\(\\ell\+1\)\}=\\sum\_\{r=0\}^\{K\}\\Phi\_\{r\}\\left\(\\tilde\{A\}^\{r\}H^\{\(\\ell\)\}\\right\)\+b\.\(5\)HereKKis the maximum hop number\. The hop\-specific Fourier KAN transform is

Φr​\(x\)o=∑i=1din∑q=1Gao​i​q\(r\)​\[cos⁡\(q​xi/s\)−1\]\+bo​i​q\(r\)​sin⁡\(q​xi/s\)\.\\Phi\_\{r\}\(x\)\_\{o\}=\\sum\_\{i=1\}^\{d\_\{\\mathrm\{in\}\}\}\\sum\_\{q=1\}^\{G\}a\_\{oiq\}^\{\(r\)\}\\left\[\\cos\(qx\_\{i\}/s\)\-1\\right\]\+b\_\{oiq\}^\{\(r\)\}\\sin\(qx\_\{i\}/s\)\.\(6\)HereGGis the number of Fourier harmonics ands=10s=10scales inputs to avoid overly rapid oscillation\. Thecos⁡\(⋅\)−1\\cos\(\\cdot\)\-1term centers the basis at zero input\. This gives TAKGCN both high\-order graph receptive fields and flexible nonlinear gene\-expression fitting\.

### II\-DRecovery\-driven Dynamic Graph Learning

After GDP\-Mask, TAKGCN encodes

Zm=fenc​\(X′,At\)\.Z\_\{m\}=f\_\{\\mathrm\{enc\}\}\(X^\{\\prime\},A\_\{t\}\)\.\(7\)The mask predictor estimates changed entries,

M^=fm​\(Zm\)\.\\hat\{M\}=f\_\{m\}\(Z\_\{m\}\)\.\(8\)with

ℒmask=BCEWithLogits​\(M^,M\)\.\\mathcal\{L\}\_\{\\mathrm\{mask\}\}=\\mathrm\{BCEWithLogits\}\(\\hat\{M\},M\)\.\(9\)The feature decoder uses bothZmZ\_\{m\}andσ​\(M^\)\\sigma\(\\hat\{M\}\):

X^=fx​\(\[Zm,σ​\(M^\)\]\)\.\\hat\{X\}=f\_\{x\}\\left\(\\left\[Z\_\{m\},\\sigma\(\\hat\{M\}\)\\right\]\\right\)\.\(10\)Weighted MSE emphasizes observed\-change positions:

ℒrec=1n​p​∑i,jwi​j​\(X^i​j−Xi​j\)2\\displaystyle\\mathcal\{L\}\_\{\\mathrm\{rec\}\}=\\frac\{1\}\{np\}\\sum\_\{i,j\}w\_\{ij\}\(\\hat\{X\}\_\{ij\}\-X\_\{ij\}\)^\{2\}\(11\)wi​j=α​Mi​j\+\(1−α\)​\(1−Mi​j\)\.\\displaystyle w\_\{ij\}=\\alpha M\_\{ij\}\+\(1\-\\alpha\)\(1\-M\_\{ij\}\)\.\(12\)The ZINB decoder provides an auxiliaryℒZINB\\mathcal\{L\}\_\{\\mathrm\{ZINB\}\}for zero inflation and overdispersion\. Crucially,X^\\hat\{X\}also drives graph learning\. During pre\-training,U=X^U=\\hat\{X\}; during refinement,U=Zc=fenc​\(X,At\)U=Z\_\{c\}=f\_\{\\mathrm\{enc\}\}\(X,A\_\{t\}\)\. After row\-normalizingUU, the model computes

Ri​j=ui⊤​uj‖ui‖2​‖uj‖2,Ri​i=−∞\.R\_\{ij\}=\\frac\{u\_\{i\}^\{\\top\}u\_\{j\}\}\{\\\|u\_\{i\}\\\|\_\{2\}\\\|u\_\{j\}\\\|\_\{2\}\},\\quad R\_\{ii\}=\-\\infty\.\(13\)With Gumbel noise

Gi​j=−log⁡\(−log⁡ϵi​j\),ϵi​j∼Uniform​\(0,1\),G\_\{ij\}=\-\\log\(\-\\log\\epsilon\_\{ij\}\),\\quad\\epsilon\_\{ij\}\\sim\\mathrm\{Uniform\}\(0,1\),\(14\)the soft edge probability is

Pi​j=softmaxj​\(\(Ri​j\+Gi​j\)/τg\),P\_\{ij\}=\\mathrm\{softmax\}\_\{j\}\\left\(\(R\_\{ij\}\+G\_\{ij\}\)/\\tau\_\{g\}\\right\),\(15\)whereτg\\tau\_\{g\}is the Gumbel temperature\. Top\-kkselection yields

Ai​jhard=𝕀​\(j∈TopK​\(Pi,k\)\)\.A^\{\\mathrm\{hard\}\}\_\{ij\}=\\mathbb\{I\}\\left\(j\\in\\mathrm\{TopK\}\(P\_\{i\},k\)\\right\)\.\(16\)The straight\-through graph keeps a sparse forward pass and soft gradients:

Pk=P⊙Ahard,P^\{k\}=P\\odot A^\{\\mathrm\{hard\}\},\(17\)Ast=Ahard−sg​\(Pk\)\+Pk,A^\{\\mathrm\{st\}\}=A^\{\\mathrm\{hard\}\}\-\\mathrm\{sg\}\(P^\{k\}\)\+P^\{k\},\(18\)wheresg​\(⋅\)\\mathrm\{sg\}\(\\cdot\)stops gradients\. After symmetrization and self\-loop removal,AdynstA\_\{\\mathrm\{dyn\}\}^\{\\mathrm\{st\}\}re\-encodes clean expression:

Zdyn=fenc​\(X,Adynst\)\.Z\_\{\\mathrm\{dyn\}\}=f\_\{\\mathrm\{enc\}\}\(X,A\_\{\\mathrm\{dyn\}\}^\{\\mathrm\{st\}\}\)\.\(19\)The hard symmetrized graph updatesAt\+1A\_\{t\+1\}after each epoch\. InfoNCE aligns current and dynamic views:

ℐ​\(Za,Zb\)=−1n​∑i=1nlog⁡exp⁡\(sim​\(zia,zib\)/τc\)∑j=1nexp⁡\(sim​\(zia,zjb\)/τc\),\\mathcal\{I\}\(Z^\{a\},Z^\{b\}\)=\-\\frac\{1\}\{n\}\\sum\_\{i=1\}^\{n\}\\log\\frac\{\\exp\(\\mathrm\{sim\}\(z\_\{i\}^\{a\},z\_\{i\}^\{b\}\)/\\tau\_\{c\}\)\}\{\\sum\_\{j=1\}^\{n\}\\exp\(\\mathrm\{sim\}\(z\_\{i\}^\{a\},z\_\{j\}^\{b\}\)/\\tau\_\{c\}\)\},\(20\)whereτc\\tau\_\{c\}is the contrastive temperature\. The contrastive losses are

ℒconp​r​e=ℐ​\(Zm,Zdyn\),\\mathcal\{L\}\_\{\\mathrm\{con\}\}^\{pre\}=\\mathcal\{I\}\(Z\_\{m\},Z\_\{\\mathrm\{dyn\}\}\),\(21\)ℒconf​i​n​e=ℐ​\(Zc,Zdyn\),\\mathcal\{L\}\_\{\\mathrm\{con\}\}^\{fine\}=\\mathcal\{I\}\(Z\_\{c\},Z\_\{\\mathrm\{dyn\}\}\),\(22\)withZc=fenc​\(X,At\)Z\_\{c\}=f\_\{\\mathrm\{enc\}\}\(X,A\_\{t\}\)\. DEC\-style refinement suppliesℒclu=KL​\(P∥Q\)\\mathcal\{L\}\_\{\\mathrm\{clu\}\}=\\mathrm\{KL\}\(P\\\|Q\)\. The full objectives are

ℒpre=λr​ℒrec\+λm​ℒmask\+λz​ℒZINB\+λc​ℒconp​r​e\.\\mathcal\{L\}\_\{\\mathrm\{pre\}\}=\\lambda\_\{r\}\\mathcal\{L\}\_\{\\mathrm\{rec\}\}\+\\lambda\_\{m\}\\mathcal\{L\}\_\{\\mathrm\{mask\}\}\+\\lambda\_\{z\}\\mathcal\{L\}\_\{\\mathrm\{ZINB\}\}\+\\lambda\_\{c\}\\mathcal\{L\}\_\{\\mathrm\{con\}\}^\{pre\}\.\(23\)ℒfine=λclu​ℒclu\+λz​ℒZINB\+λc​ℒconf​i​n​e\.\\mathcal\{L\}\_\{\\mathrm\{fine\}\}=\\lambda\_\{\\mathrm\{clu\}\}\\mathcal\{L\}\_\{\\mathrm\{clu\}\}\+\\lambda\_\{z\}\\mathcal\{L\}\_\{\\mathrm\{ZINB\}\}\+\\lambda\_\{c\}\\mathcal\{L\}\_\{\\mathrm\{con\}\}^\{fine\}\.\(24\)

## IIIExperiments

### III\-ADatasets and Evaluation Protocol

We evaluate scKDGM on 12 real scRNA\-seq datasets from Adam\[[1](https://arxiv.org/html/2606.28459#bib.bib29)\], Klein\[[24](https://arxiv.org/html/2606.28459#bib.bib30)\], Plasschaert\[[41](https://arxiv.org/html/2606.28459#bib.bib31)\], Tabula Muris\[[51](https://arxiv.org/html/2606.28459#bib.bib32)\], Romanov\[[46](https://arxiv.org/html/2606.28459#bib.bib33)\], Wang Lung\[[62](https://arxiv.org/html/2606.28459#bib.bib34)\], and Young\[[82](https://arxiv.org/html/2606.28459#bib.bib35)\]\. They cover kidney, embryonic stem cells, trachea, limb muscle, diaphragm, heart, hypothalamus, and lung; Drop\-seq, inDrop, 10x, and Smart\-seq2; 2 to 11 cell types; 870 to 11269 cells; and 65\.58% to 94\.70% zero entries\. We compare 10 baselines: graph\-based scMGCA\[[85](https://arxiv.org/html/2606.28459#bib.bib18)\], scAGC\[[25](https://arxiv.org/html/2606.28459#bib.bib21)\], scCDCG\[[74](https://arxiv.org/html/2606.28459#bib.bib19)\], scDSC\[[12](https://arxiv.org/html/2606.28459#bib.bib16)\], and scGNN\[[56](https://arxiv.org/html/2606.28459#bib.bib15)\], plus CIRCLE\[[55](https://arxiv.org/html/2606.28459#bib.bib13)\], scDML\[[83](https://arxiv.org/html/2606.28459#bib.bib11)\], scziDesk\[[7](https://arxiv.org/html/2606.28459#bib.bib10)\], scDeepCluster\[[53](https://arxiv.org/html/2606.28459#bib.bib9)\], and scMAE\[[11](https://arxiv.org/html/2606.28459#bib.bib12)\]\. Metrics are NMI and ARI\.

### III\-BImplementation Details

All experiments are implemented in PyTorch and executed on an NVIDIA A6000 GPU with 48 GB memory\. scKDGM uses a hidden dimension of 256, a latent dimension of 128, an encoder dropout rate of 0\.2, and TAKGConv withK=3K=3and Fourier harmonic numberG=4G=4\. Pre\-training runs for 1000 epochs with a learning rate of10−410^\{\-4\}\. GDP\-Mask uses a mask rate of 0\.3, the dynamic graph neighbor number isk=15k=15, the Gumbel temperature is 1\.0, the InfoNCE temperature is 0\.7, and the loss weights areλr=1\.0,λm=0\.1,λz=1\.0,λc=0\.1\\lambda\_\{r\}=1\.0,\\lambda\_\{m\}=0\.1,\\lambda\_\{z\}=1\.0,\\lambda\_\{c\}=0\.1\. Clustering refinement runs for 200 epochs withλclu=1\.0,λz=0\.1,λc=0\.01\\lambda\_\{\\mathrm\{clu\}\}=1\.0,\\lambda\_\{z\}=0\.1,\\lambda\_\{c\}=0\.01, and the target distribution is updated every 8 epochs\. Ablation and parameter sensitivity experiments follow the same training protocol and change only the module or parameter under analysis\.

### III\-COverall Clustering Performance

Table[I](https://arxiv.org/html/2606.28459#S3.T1)reports mean scores over three seeds against ground\-truth labels\. Baselines marked with∗use unified reproduced results from scCluBench\[[75](https://arxiv.org/html/2606.28459#bib.bib28)\]; other baselines follow the same protocol\. scKDGM achieves the best average NMI/ARI of 0\.8854/0\.9105, ranking first on 9 of 12 datasets for NMI and 8 of 12 for ARI\.

TABLE I:Overall clustering performance on 12 scRNA\-seq datasets\.MetricDatasetOursGNN MethodsDeep MethodsscKDGMscMGCAscAGCscCDCG\*scDSC\*scGNN\*CIRCLEscDMLscziDeskscDeepCluster\*scMAE\*NMIAdam0\.9381\\mathbf\{0\.9381\}0\.86640\.86640\.8871¯\\underline\{0\.8871\}0\.62460\.62460\.78890\.78890\.71290\.71290\.69990\.69990\.85470\.85470\.82750\.82750\.64660\.64660\.79010\.7901Klein0\.8245\\mathbf\{0\.8245\}0\.69390\.69390\.80040\.80040\.77530\.77530\.57480\.57480\.64920\.64920\.76520\.76520\.59570\.59570\.78550\.78550\.73230\.73230\.8134¯\\underline\{0\.8134\}Plasschaert0\.8438¯\\underline\{0\.8438\}0\.74570\.74570\.8582\\mathbf\{0\.8582\}0\.65590\.65590\.76900\.76900\.54190\.54190\.64000\.64000\.67930\.67930\.79520\.79520\.60470\.60470\.71980\.7198Qx\_Limb\_Muscle0\.9832\\mathbf\{0\.9832\}0\.94170\.94170\.94020\.94020\.94120\.94120\.79700\.79700\.77600\.77600\.90500\.90500\.94700\.94700\.89870\.89870\.91120\.91120\.9682¯\\underline\{0\.9682\}Qx\_Trachea0\.8788\\mathbf\{0\.8788\}0\.70180\.70180\.83350\.83350\.49840\.49840\.67500\.67500\.36360\.36360\.57360\.57360\.67160\.67160\.8473¯\\underline\{0\.8473\}0\.52540\.52540\.83140\.8314QS\_Diaphragm0\.9765\\mathbf\{0\.9765\}0\.93500\.93500\.94730\.94730\.84310\.84310\.91820\.91820\.94460\.94460\.85440\.85440\.85590\.85590\.94290\.94290\.90660\.90660\.9522¯\\underline\{0\.9522\}QS\_Heart0\.9199\\mathbf\{0\.9199\}0\.90920\.90920\.9106¯\\underline\{0\.9106\}0\.84650\.84650\.86810\.86810\.68140\.68140\.76710\.76710\.84650\.84650\.83340\.83340\.80700\.80700\.86440\.8644QS\_Limb\_Muscle0\.9637¯\\underline\{0\.9637\}0\.93970\.93970\.9645\\mathbf\{0\.9645\}0\.87790\.87790\.88470\.88470\.87360\.87360\.77090\.77090\.94700\.94700\.95870\.95870\.89460\.89460\.95110\.9511QS\_Trachea0\.7528¯\\underline\{0\.7528\}0\.7710\\mathbf\{0\.7710\}0\.68980\.68980\.66340\.66340\.56870\.56870\.74420\.74420\.57150\.57150\.63750\.63750\.64680\.64680\.64930\.64930\.64140\.6414Romanov0\.7800\\mathbf\{0\.7800\}0\.63080\.63080\.66540\.66540\.60180\.60180\.54630\.54630\.55770\.55770\.66540\.66540\.52040\.52040\.7278¯\\underline\{0\.7278\}0\.57550\.57550\.69310\.6931Wang\_Lung0\.9231\\mathbf\{0\.9231\}0\.65160\.65160\.87860\.87860\.83470\.83470\.85260\.85260\.82520\.82520\.85710\.85710\.81770\.81770\.81920\.81920\.67480\.67480\.9184¯\\underline\{0\.9184\}Young0\.8406\\mathbf\{0\.8406\}0\.80460\.80460\.78220\.78220\.61480\.61480\.72830\.72830\.40470\.40470\.75650\.75650\.8210¯\\underline\{0\.8210\}0\.75370\.75370\.49310\.49310\.67900\.6790AVG0\.8854\\mathbf\{0\.8854\}0\.79930\.79930\.8465¯\\underline\{0\.8465\}0\.73150\.73150\.74760\.74760\.67290\.67290\.73550\.73550\.76620\.76620\.81970\.81970\.70180\.70180\.81850\.8185ARIAdam0\.9502\\mathbf\{0\.9502\}0\.87250\.87250\.9028¯\\underline\{0\.9028\}0\.45370\.45370\.74840\.74840\.59660\.59660\.62980\.62980\.85060\.85060\.80330\.80330\.45750\.45750\.73780\.7378Klein0\.8174¯\\underline\{0\.8174\}0\.68860\.68860\.8188\\mathbf\{0\.8188\}0\.75330\.75330\.57280\.57280\.57710\.57710\.77230\.77230\.61280\.61280\.80960\.80960\.70100\.70100\.75950\.7595Plasschaert0\.9017¯\\underline\{0\.9017\}0\.77180\.77180\.9144\\mathbf\{0\.9144\}0\.63990\.63990\.83200\.83200\.41250\.41250\.48370\.48370\.65630\.65630\.84950\.84950\.42550\.42550\.61620\.6162Qx\_Limb\_Muscle0\.9902\\mathbf\{0\.9902\}0\.95720\.95720\.95250\.95250\.96600\.96600\.78200\.78200\.75410\.75410\.86950\.86950\.94480\.94480\.90590\.90590\.93320\.93320\.9836¯\\underline\{0\.9836\}Qx\_Trachea0\.9670\\mathbf\{0\.9670\}0\.54040\.54040\.93400\.93400\.46590\.46590\.72780\.72780\.30020\.30020\.36210\.36210\.66830\.66830\.92010\.92010\.37300\.37300\.9456¯\\underline\{0\.9456\}QS\_Diaphragm0\.9872\\mathbf\{0\.9872\}0\.96120\.96120\.9747¯\\underline\{0\.9747\}0\.91840\.91840\.94160\.94160\.96990\.96990\.85500\.85500\.82720\.82720\.96680\.96680\.93530\.93530\.97140\.9714QS\_Heart0\.9615\\mathbf\{0\.9615\}0\.95210\.95210\.9580¯\\underline\{0\.9580\}0\.91620\.91620\.92600\.92600\.55440\.55440\.61310\.61310\.83030\.83030\.75440\.75440\.69280\.69280\.77570\.7757QS\_Limb\_Muscle0\.9808¯\\underline\{0\.9808\}0\.97010\.97010\.9816\\mathbf\{0\.9816\}0\.93150\.93150\.93050\.93050\.88330\.88330\.63020\.63020\.96690\.96690\.97230\.97230\.90970\.90970\.97180\.9718QS\_Trachea0\.8396¯\\underline\{0\.8396\}0\.8686\\mathbf\{0\.8686\}0\.81330\.81330\.59950\.59950\.54510\.54510\.78070\.78070\.41030\.41030\.55280\.55280\.70650\.70650\.48500\.48500\.53890\.5389Romanov0\.7969\\mathbf\{0\.7969\}0\.53030\.53030\.63300\.63300\.58610\.58610\.41750\.41750\.51400\.51400\.61050\.61050\.40630\.40630\.7507¯\\underline\{0\.7507\}0\.47950\.47950\.67050\.6705Wang\_Lung0\.9690\\mathbf\{0\.9690\}0\.72880\.72880\.93940\.93940\.91340\.91340\.92570\.92570\.89750\.89750\.93200\.93200\.89880\.89880\.90230\.90230\.66370\.66370\.9667¯\\underline\{0\.9667\}Young0\.7644\\mathbf\{0\.7644\}0\.70780\.70780\.69460\.69460\.41660\.41660\.66750\.66750\.19070\.19070\.70850\.70850\.7163¯\\underline\{0\.7163\}0\.64480\.64480\.26180\.26180\.51620\.5162AVG0\.9105\\mathbf\{0\.9105\}0\.79580\.79580\.8764¯\\underline\{0\.8764\}0\.71340\.71340\.75140\.75140\.61930\.61930\.65640\.65640\.74430\.74430\.83220\.83220\.60980\.60980\.78780\.7878
### III\-DAblation Study

We compare full scKDGM with three variants\. w/o Mask removes GDP\-Mask pre\-training and recovery supervision; w/o KAN replaces TAKGConv with a comparable TAGCN encoder of about 9M parameters; w/o DG disables dynamic graph updates\. Fig\.[2](https://arxiv.org/html/2606.28459#S3.F2)shows that w/o DG drops average NMI/ARI by 0\.0919/0\.1309, confirming the value of recovery\-driven graph updates for reducing information bottlenecks, uneven propagation, and oversmoothing in static KNN graphs\. w/o Mask drops NMI/ARI by 0\.0596/0\.0804, supporting graph\-aware perturbation and observed\-change recovery\. w/o KAN also declines, showing that Fourier KAN transformations improve nonlinear high\-order neighborhood aggregation\.

![Refer to caption](https://arxiv.org/html/2606.28459v1/x2.png)Figure 2:Ablation and parameter sensitivity results\. Panel \(a\) reports average performance changes relative to full scKDGM, where the TAGCN variant uses an encoder with a comparable parameter count\. Panels \(b\)\-\(d\) report sensitivity to mask rate, graph neighbor numberkk, and Fourier harmonic numberGGon Quake Smart\-seq2 Diaphragm\.
### III\-EDynamic Graph Study

We further inspect cell type consistency and degree distribution\. Edge homophily is the fraction of edges linking cells of the same type\. In Fig\.[3](https://arxiv.org/html/2606.28459#S3.F3)a, dynamic\-graph homophily rises during training, reaches about 0\.599 in refinement, and peaks at about 0\.611, indicating increasingly cell\-type\-consistent edges\. Fig\.[3](https://arxiv.org/html/2606.28459#S3.F3)b shows that the KNN graph has a long\-tailed hub structure, with maximum degree 175 and 99% quantile 96\.3\. The dynamic graph reduces them to 44 and 37\.0, suppressing hub concentration and mitigating uneven propagation, semantic mixing, and oversmoothing\.

![Refer to caption](https://arxiv.org/html/2606.28459v1/x3.png)Figure 3:Graph\-structure diagnostics on the Quake Smart\-seq2 Diaphragm dataset\. \(a\) Training evolution of dynamic graph edge homophily\. \(b\) Degree\-tail distributions of the KNN\-constructed graph and the learned dynamic graph\. The two graphs are constructed with the same neighbor numberkk\.
### III\-FParameter Sensitivity

As shown in Fig\.[2](https://arxiv.org/html/2606.28459#S3.F2), the mask rate reaches the best NMI/ARI at 0\.3 on Quake Smart\-seq2 Diaphragm, with 0\.9773/0\.9893, and remains stable from 0\.3 to 0\.5\. Thus moderate perturbation supplies useful self\-supervision, while excessive masking weakens cell\-specific patterns\. For the dynamic graph,k=15k=15gives 0\.9731/0\.9875; smallerkklacks coverage, and largerkkincreases semantic mixing\. For Fourier harmonics,G=4G=4is best with 0\.9731/0\.9875, balancing basis capacity and stability\. The defaultsk=15k=15,G=4G=4, and mask rate 0\.3 therefore lie in the stable high\-performance range\.

## IVConclusion

This paper presents scKDGM, a KAN\-guided dynamic graph masked learning framework for scRNA\-seq clustering\. GDP\-Mask builds graph\-aware distribution preserving perturbations, TAKGCN learns nonlinear graph\-aware representations, and the mask\-recovered feature matrix updates the dynamic graph, closing the loop between expression recovery and topology optimization\. On 12 real datasets, scKDGM achieves higher average NMI and ARI than representative deep and graph clustering methods\. Ablation and sensitivity studies confirm the contributions of dynamic graph updates, GDP\-Mask, and TAKGConv\. However, pairwise dynamic graph construction needs approximate neighbor search or mini\-batch updates for million\-cell data, and biological interpretation should be further validated by marker enrichment and cell type annotation consistency\. Future work will improve graph efficiency and extend scKDGM to batch correction, multi\-omics clustering, and spatial transcriptomics\.

## Acknowledgment

This work was supported in part by the National Key R&D Program of China under Grant 2025YFC3409000\.

## References

- \[1\]M\. Adam, A\. S\. Potter, and S\. S\. Potter\(2017\)Psychrophilic proteases dramatically reduce single\-cell rna\-seq artifacts: a molecular atlas of kidney development\.Development144\(19\),pp\. 3625–3632\.Cited by:[§III\-A](https://arxiv.org/html/2606.28459#S3.SS1.p1.1)\.
- \[2\]F\. Bi, T\. He, and X\. Luo\(2024\-05\)A fast nonnegative autoencoder\-based approach to latent feature analysis on high\-dimensional and incomplete data\.IEEE Transactions on Services Computing17\(3\),pp\. 733–746\.External Links:ISSN 2372\-0204,[Link](http://dx.doi.org/10.1109/tsc.2023.3319713),[Document](https://dx.doi.org/10.1109/tsc.2023.3319713)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p3.1)\.
- \[3\]F\. Bi, T\. He, Y\. Ong, and X\. Luo\(2025\-11\)Discovering spatiotemporal–individual coupled features from nonstandard tensors—a novel dynamic graph mixer approach\.IEEE Transactions on Neural Networks and Learning Systems36\(11\),pp\. 19834–19848\.External Links:ISSN 2162\-2388,[Link](http://dx.doi.org/10.1109/tnnls.2025.3592692),[Document](https://dx.doi.org/10.1109/tnnls.2025.3592692)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p4.1)\.
- \[4\]F\. Bi, T\. He, Y\. Ong, and X\. Luo\(2025\-04\)Graph linear convolution pooling for learning in incomplete high\-dimensional data\.IEEE Transactions on Knowledge and Data Engineering37\(4\),pp\. 1838–1852\.External Links:ISSN 2326\-3865,[Link](http://dx.doi.org/10.1109/tkde.2024.3524627),[Document](https://dx.doi.org/10.1109/tkde.2024.3524627)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p2.1)\.
- \[5\]R\. Bresson, G\. Nikolentzos, G\. Panagopoulos, M\. Chatzianastasis, J\. Pang, and M\. Vazirgiannis\(2024\)Kagnns: kolmogorov\-arnold networks meet graph learning\.arXiv preprint arXiv:2406\.18380\.Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p5.1)\.
- \[6\]X\. Cao, Y\. Huang, Z\. You, X\. Shang, L\. Hu, P\. Hu, and Z\. Huang\(2024\)ScPriorGraph: constructing biosemantic cell–cell graphs with prior gene set selection for cell type identification from scrna\-seq data\.Genome Biology25\(1\),pp\. 207\.Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p2.1)\.
- \[7\]L\. Chen, W\. Wang, Y\. Zhai, and M\. Deng\(2020\-06\)Deep soft k\-means clustering with self\-training for single\-cell rna sequence data\.NAR Genomics and Bioinformatics2\(2\),pp\. lqaa039\.External Links:ISSN 2631\-9268,[Document](https://dx.doi.org/10.1093/nargab/lqaa039),[Link](https://doi.org/10.1093/nargab/lqaa039),https://academic\.oup\.com/nargab/article\-pdf/2/2/lqaa039/34054328/lqaa039\.pdfCited by:[§I](https://arxiv.org/html/2606.28459#S1.p2.1),[§III\-A](https://arxiv.org/html/2606.28459#S3.SS1.p1.1)\.
- \[8\]M\. Chen, L\. Tao, J\. Lou, and X\. Luo\(2025\-03\)Latent\-factorization\-of\-tensors\-incorporated battery cycle life prediction\.IEEE/CAA Journal of Automatica Sinica12\(3\),pp\. 633–635\.External Links:ISSN 2329\-9274,[Link](http://dx.doi.org/10.1109/jas.2024.124602),[Document](https://dx.doi.org/10.1109/jas.2024.124602)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p4.1)\.
- \[9\]X\. Deng, P\. Hu, T\. Herget, F\. Tan, X\. Zhu, J\. Zhang, Y\. Huang, L\. Hu, Z\. You, and X\. Luo\(2026\-01\)Fuzzy mixture\-of\-experts aggregation for organoid identification with multiscale state space features\.IEEE Transactions on Fuzzy Systems34\(1\),pp\. 324–335\.External Links:ISSN 1941\-0034,[Link](http://dx.doi.org/10.1109/tfuzz.2025.3622935),[Document](https://dx.doi.org/10.1109/tfuzz.2025.3622935)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p1.1)\.
- \[10\]J\. Du, S\. Zhang, G\. Wu, J\. M\. Moura, and S\. Kar\(2017\)Topology adaptive graph convolutional networks\.arXiv preprint arXiv:1710\.10370\.Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p5.1)\.
- \[11\]Z\. Fang, R\. Zheng, and M\. Li\(2024\)ScMAE: a masked autoencoder for single\-cell rna\-seq clustering\.Bioinformatics40\(1\),pp\. btae020\.Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p2.1),[§III\-A](https://arxiv.org/html/2606.28459#S3.SS1.p1.1)\.
- \[12\]Y\. Gan, X\. Huang, G\. Zou, S\. Zhou, and J\. Guan\(2022\)Deep structural clustering for single\-cell rna\-seq data jointly through autoencoder and graph neural network\.Briefings in Bioinformatics23\(2\),pp\. bbac018\.Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p2.1),[§III\-A](https://arxiv.org/html/2606.28459#S3.SS1.p1.1)\.
- \[13\]J\. Gou, Y\. Cheng, B\. Ma, L\. Du, X\. Luo, and Z\. Yi\(2026\-01\)Multi\-scale collaborative distillation graph neural networks for session\-based recommendation\.IEEE Transactions on Services Computing19\(1\),pp\. 504–517\.External Links:ISSN 2372\-0204,[Link](http://dx.doi.org/10.1109/tsc.2025.3637009),[Document](https://dx.doi.org/10.1109/tsc.2025.3637009)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p2.1)\.
- \[14\]M\. Han, L\. Wang, Y\. Yuan, and X\. Luo\(2025\-08\)SGD\-dyg: self\-reliant global dependency apprehending on dynamic graphs\.InProceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V\.2,KDD ’25,pp\. 802–813\.External Links:[Link](http://dx.doi.org/10.1145/3711896.3737126),[Document](https://dx.doi.org/10.1145/3711896.3737126)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p2.1)\.
- \[15\]T\. He, Z\. Duan, and X\. Luo\(2026\-03\)Modularized graph convolutional network\.IEEE/CAA Journal of Automatica Sinica13\(3\),pp\. 737–739\.External Links:ISSN 2329\-9274,[Link](http://dx.doi.org/10.1109/jas.2025.125336),[Document](https://dx.doi.org/10.1109/jas.2025.125336)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p5.1)\.
- \[16\]Y\. He and X\. Luo\(2026\-01\)Tensor low\-rank orthogonal compression for convolutional neural networks\.IEEE/CAA Journal of Automatica Sinica13\(1\),pp\. 227–229\.External Links:ISSN 2329\-9274,[Link](http://dx.doi.org/10.1109/jas.2025.125213),[Document](https://dx.doi.org/10.1109/jas.2025.125213)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p5.1)\.
- \[17\]Z\. He, M\. Lin, X\. Luo, and Z\. Xu\(2025\-07\)Structure\-preserved self\-attention for fusion image information in multiple color spaces\.IEEE Transactions on Neural Networks and Learning Systems36\(7\),pp\. 13021–13035\.External Links:ISSN 2162\-2388,[Link](http://dx.doi.org/10.1109/tnnls.2024.3490800),[Document](https://dx.doi.org/10.1109/tnnls.2024.3490800)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p1.1)\.
- \[18\]Y\. Hou, P\. Tang, and X\. Luo\(2026\-04\)Multi\-aspect self\-attending neural tucker factorization for spatiotemporal representation learning\.IEEE/CAA Journal of Automatica Sinica13\(4\),pp\. 986–988\.External Links:ISSN 2329\-9274,[Link](http://dx.doi.org/10.1109/jas.2025.125723),[Document](https://dx.doi.org/10.1109/jas.2025.125723)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p4.1)\.
- \[19\]Z\. Hou, X\. Liu, Y\. Cen, Y\. Dong, H\. Yang, C\. Wang, and J\. Tang\(2022\)Graphmae: self\-supervised masked graph autoencoders\.InProceedings of the 28th ACM SIGKDD conference on knowledge discovery and data mining,pp\. 594–604\.Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p2.1)\.
- \[20\]Q\. Hu, H\. Wu, and X\. Luo\(2025\-12\)A comprehensive review of parallel optimization algorithms for high\-dimensional and incomplete matrix factorization\.IEEE/CAA Journal of Automatica Sinica12\(12\),pp\. 2399–2426\.External Links:ISSN 2329\-9274,[Link](http://dx.doi.org/10.1109/jas.2025.125774),[Document](https://dx.doi.org/10.1109/jas.2025.125774)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p3.1)\.
- \[21\]Y\. Huang, Y\. Li, Z\. You, L\. Hu, P\. Hu, L\. Wang, Y\. Peng, and Z\. Huang\(2025\)Consensus representation of multiple cell–cell graphs from gene signaling pathways for cell type annotation\.BMC biology23\(1\),pp\. 23\.Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p2.1)\.
- \[22\]V\. Y\. Kiselev, T\. S\. Andrews, and M\. Hemberg\(2019\)Challenges in unsupervised clustering of single\-cell rna\-seq data\.Nature Reviews Genetics20\(5\),pp\. 273–282\.Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p1.1)\.
- \[23\]V\. Y\. Kiselev, K\. Kirschner, M\. T\. Schaub, T\. Andrews, A\. Yiu, T\. Chandra, K\. N\. Natarajan, W\. Reik, M\. Barahona, A\. R\. Green,et al\.\(2017\)SC3: consensus clustering of single\-cell rna\-seq data\.Nature methods14\(5\),pp\. 483–486\.Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p2.1)\.
- \[24\]A\. M\. Klein, L\. Mazutis, I\. Akartuna, N\. Tallapragada, A\. Veres, V\. Li, L\. Peshkin, D\. A\. Weitz, and M\. W\. Kirschner\(2015\)Droplet barcoding for single\-cell transcriptomics applied to embryonic stem cells\.Cell161\(5\),pp\. 1187–1201\.Cited by:[§III\-A](https://arxiv.org/html/2606.28459#S3.SS1.p1.1)\.
- \[25\]H\. Li, J\. Fu, X\. Zhuang, H\. Yang, X\. Ling, T\. Cheng, I\. Razzak, Z\. Chen,et al\.\(2025\)ScAGC: learning adaptive cell graphs with contrastive guidance for single\-cell clustering\.arXiv preprint arXiv:2508\.09180\.Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p2.1),[§III\-A](https://arxiv.org/html/2606.28459#S3.SS1.p1.1)\.
- \[26\]J\. Li, Y\. Yuan, T\. He, and X\. Luo\(2026\-07\)Adaptive pid\-incorporated nonnegative latent factor analysis\.IEEE Transactions on Systems, Man, and Cybernetics: Systems56\(7\),pp\. 4384–4397\.External Links:ISSN 2168\-2232,[Link](http://dx.doi.org/10.1109/tsmc.2026.3678292),[Document](https://dx.doi.org/10.1109/tsmc.2026.3678292)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p3.1)\.
- \[27\]J\. Li, Y\. Yuan, and X\. Luo\(2025\-10\)Learning error refinement in stochastic gradient descent\-based latent factor analysis via diversified pid controllers\.IEEE Transactions on Emerging Topics in Computational Intelligence9\(5\),pp\. 3582–3597\.External Links:ISSN 2471\-285X,[Link](http://dx.doi.org/10.1109/tetci.2025.3547854),[Document](https://dx.doi.org/10.1109/tetci.2025.3547854)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p3.1)\.
- \[28\]L\. Li, Y\. Zhang, G\. Wang, and K\. Xia\(2025\)Kolmogorov–arnold graph neural networks for molecular property prediction\.Nature Machine Intelligence7\(8\),pp\. 1346–1354\.Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p5.1)\.
- \[29\]W\. Li, R\. Wang, and X\. Luo\(2025\-01\)A generalized nesterov\-accelerated second\-order latent factor model for high\-dimensional and incomplete data\.IEEE Transactions on Neural Networks and Learning Systems36\(1\),pp\. 1518–1532\.External Links:ISSN 2162\-2388,[Link](http://dx.doi.org/10.1109/tnnls.2023.3321915),[Document](https://dx.doi.org/10.1109/tnnls.2023.3321915)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p3.1)\.
- \[30\]W\. Li, M\. Lin, X\. Xu, L\. Lin, Z\. Xu, and X\. Luo\(2026\-01\)Neural nonnegative latent factorization of tensors model with acceleration and unconstraint\.IEEE Transactions on Systems, Man, and Cybernetics: Systems56\(1\),pp\. 164–178\.External Links:ISSN 2168\-2232,[Link](http://dx.doi.org/10.1109/tsmc.2025.3622727),[Document](https://dx.doi.org/10.1109/tsmc.2025.3622727)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p4.1)\.
- \[31\]X\. Liao, H\. Wu, T\. He, and X\. Luo\(2025\-11\)A proximal\-admm\-incorporated nonnegative latent\-factorization\-of\-tensors model for representing dynamic cryptocurrency transaction network\.IEEE Transactions on Systems, Man, and Cybernetics: Systems55\(11\),pp\. 8387–8401\.External Links:ISSN 2168\-2232,[Link](http://dx.doi.org/10.1109/tsmc.2025.3605054),[Document](https://dx.doi.org/10.1109/tsmc.2025.3605054)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p4.1)\.
- \[32\]X\. Liao, H\. Wu, and X\. Luo\(2025\)A novel tensor causal convolution network model for highly\-accurate representation to spatio\-temporal data\.IEEE Transactions on Automation Science and Engineering22,pp\. 19525–19537\.External Links:ISSN 1558\-3783,[Link](http://dx.doi.org/10.1109/tase.2025.3595545),[Document](https://dx.doi.org/10.1109/tase.2025.3595545)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p4.1)\.
- \[33\]L\. Lin, Q\. Li, M\. Qiao, Z\. Wang, J\. Zhao, R\. Li, X\. Luo, and T\. Jia\(2026\-02\)NCSAC: effective neural community search via attribute\-augmented conductance\.IEEE Transactions on Knowledge and Data Engineering38\(2\),pp\. 1221–1235\.External Links:ISSN 2326\-3865,[Link](http://dx.doi.org/10.1109/tkde.2025.3630626),[Document](https://dx.doi.org/10.1109/tkde.2025.3630626)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p2.1)\.
- \[34\]P\. Lin, M\. Troup, and J\. W\. Ho\(2017\)CIDR: ultrafast and accurate clustering through imputation for single\-cell rna\-seq data\.Genome biology18\(1\),pp\. 59\.Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p2.1)\.
- \[35\]Z\. Liu, X\. Luo, and M\. Zhou\(2024\-04\)Symmetry and graph bi\-regularized non\-negative matrix factorization for precise community detection\.IEEE Transactions on Automation Science and Engineering21\(2\),pp\. 1406–1420\.External Links:ISSN 1558\-3783,[Link](http://dx.doi.org/10.1109/tase.2023.3240335),[Document](https://dx.doi.org/10.1109/tase.2023.3240335)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p2.1)\.
- \[36\]Z\. Liu, Y\. Wang, S\. Vaidya, F\. Ruehle, J\. Halverson, M\. Soljacic, T\. Hou, and M\. Tegmark\(2025\)KAN: kolmogorov–arnold networks\.InInternational conference on learning representations,Vol\.2025,pp\. 70367–70413\.Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p5.1)\.
- \[37\]M\. D\. Luecken and F\. J\. Theis\(2019\)Current best practices in single\-cell rna\-seq analysis: a tutorial\.Molecular systems biology15\(6\),pp\. MSB188746\.Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p1.1)\.
- \[38\]C\. Lyu, J\. Cheng, X\. Luo, and Y\. Shi\(2026\-05\)Genetic algorithm\-based two\-step optimization for precise latent factor analysis\.IEEE Transactions on Neural Networks and Learning Systems37\(5\),pp\. 2294–2306\.External Links:ISSN 2162\-2388,[Link](http://dx.doi.org/10.1109/tnnls.2025.3631465),[Document](https://dx.doi.org/10.1109/tnnls.2025.3631465)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p3.1)\.
- \[39\]C\. Lyu, Z\. Ma, X\. Luo, and Y\. Shi\(2026\-01\)Dynamic stochastic reorientation particle swarm optimization for adaptive latent factor analysis in high\-dimensional sparse matrices\.IEEE Transactions on Knowledge and Data Engineering38\(1\),pp\. 222–234\.External Links:ISSN 2326\-3865,[Link](http://dx.doi.org/10.1109/tkde.2025.3621469),[Document](https://dx.doi.org/10.1109/tkde.2025.3621469)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p3.1)\.
- \[40\]E\. Z\. Macosko, A\. Basu, R\. Satija, J\. Nemesh, K\. Shekhar, M\. Goldman, I\. Tirosh, A\. R\. Bialas, N\. Kamitaki, E\. M\. Martersteck,et al\.\(2015\)Highly parallel genome\-wide expression profiling of individual cells using nanoliter droplets\.Cell161\(5\),pp\. 1202–1214\.Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p1.1)\.
- \[41\]L\. W\. Plasschaert, R\. Žilionis, R\. Choo\-Wing, V\. Savova, J\. Knehr, G\. Roma, A\. M\. Klein, and A\. B\. Jaffe\(2018\)A single\-cell atlas of the airway epithelium reveals the cftr\-rich pulmonary ionocyte\.Nature560\(7718\),pp\. 377–381\.Cited by:[§III\-A](https://arxiv.org/html/2606.28459#S3.SS1.p1.1)\.
- \[42\]W\. Qin, Y\. Ding, and X\. Luo\(2026\-05\)A robust approach to electricity theft detection via tensor representation\-driven contrastive distillation\.IEEE Transactions on Industrial Informatics22\(5\),pp\. 4561–4570\.External Links:ISSN 1941\-0050,[Link](http://dx.doi.org/10.1109/tii.2026.3659333),[Document](https://dx.doi.org/10.1109/tii.2026.3659333)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p4.1)\.
- \[43\]W\. Qin, X\. Luo, S\. Li, and M\. Zhou\(2024\-07\)Parallel adaptive stochastic gradient descent algorithms for latent factor analysis of high\-dimensional and incomplete industrial data\.IEEE Transactions on Automation Science and Engineering21\(3\),pp\. 2716–2729\.External Links:ISSN 1558\-3783,[Link](http://dx.doi.org/10.1109/tase.2023.3267609),[Document](https://dx.doi.org/10.1109/tase.2023.3267609)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p3.1)\.
- \[44\]W\. Qin, X\. Luo, and M\. Zhou\(2024\-02\)Adaptively\-accelerated parallel stochastic gradient descent for high\-dimensional and incomplete data representation learning\.IEEE Transactions on Big Data10\(1\),pp\. 92–107\.External Links:ISSN 2372\-2096,[Link](http://dx.doi.org/10.1109/tbdata.2023.3326304),[Document](https://dx.doi.org/10.1109/tbdata.2023.3326304)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p3.1)\.
- \[45\]W\. Qin and X\. Luo\(2024\-02\)Asynchronous parallel fuzzy stochastic gradient descent for high\-dimensional incomplete data representation\.IEEE Transactions on Fuzzy Systems32\(2\),pp\. 445–459\.External Links:ISSN 1941\-0034,[Link](http://dx.doi.org/10.1109/tfuzz.2023.3300370),[Document](https://dx.doi.org/10.1109/tfuzz.2023.3300370)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p3.1)\.
- \[46\]R\. A\. Romanov, A\. Zeisel, J\. Bakker, F\. Girach, A\. Hellysaz, R\. Tomer, A\. Alpar, J\. Mulder, F\. Clotman, E\. Keimpema,et al\.\(2017\)Molecular interrogation of hypothalamic organization reveals distinct dopamine neuronal subtypes\.Nature neuroscience20\(2\),pp\. 176–188\.Cited by:[§III\-A](https://arxiv.org/html/2606.28459#S3.SS1.p1.1)\.
- \[47\]R\. Satija, J\. A\. Farrell, D\. Gennert, A\. F\. Schier, and A\. Regev\(2015\)Spatial reconstruction of single\-cell gene expression data\.Nature biotechnology33\(5\),pp\. 495–502\.Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p2.1)\.
- \[48\]F\. Tang, C\. Barbacioru, Y\. Wang, E\. Nordman, C\. Lee, N\. Xu, X\. Wang, J\. Bodeau, B\. B\. Tuch, A\. Siddiqui,et al\.\(2009\)MRNA\-seq whole\-transcriptome analysis of a single cell\.Nature methods6\(5\),pp\. 377–382\.Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p1.1)\.
- \[49\]P\. Tang, X\. Luo, and J\. Woodcock\(2025\-10\)Auto\-encoding neural tucker factorization\.IEEE Transactions on Knowledge and Data Engineering37\(10\),pp\. 5795–5807\.External Links:ISSN 2326\-3865,[Link](http://dx.doi.org/10.1109/tkde.2025.3590198),[Document](https://dx.doi.org/10.1109/tkde.2025.3590198)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p4.1)\.
- \[50\]P\. Tang and X\. Luo\(2025\-02\)Neural tucker factorization\.IEEE/CAA Journal of Automatica Sinica12\(2\),pp\. 475–477\.External Links:ISSN 2329\-9274,[Link](http://dx.doi.org/10.1109/jas.2024.124977),[Document](https://dx.doi.org/10.1109/jas.2024.124977)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p4.1)\.
- \[51\]The Tabula Muris Consortium\(2018\)Single\-cell transcriptomics of 20 mouse organs creates a tabula muris\.Nature562\(7727\),pp\. 367–372\.External Links:[Document](https://dx.doi.org/10.1038/s41586-018-0590-4)Cited by:[§III\-A](https://arxiv.org/html/2606.28459#S3.SS1.p1.1)\.
- \[52\]H\. Tian, X\. Kong, J\. Liu, J\. Shang, J\. Wang, and L\. Dai\(2025\)ScGZDC: graph\-based zinb deep clustering for single\-cell rna\-seq data\.In2025 IEEE International Conference on Bioinformatics and Biomedicine \(BIBM\),pp\. 6369–6375\.Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p2.1)\.
- \[53\]T\. Tian, J\. Wan, Q\. Song, and Z\. Wei\(2019\)Clustering single\-cell rna\-seq data with a model\-based deep learning approach\.Nature Machine Intelligence1\(4\),pp\. 191–198\.Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p2.1),[§III\-A](https://arxiv.org/html/2606.28459#S3.SS1.p1.1)\.
- \[54\]H\. Wan, L\. Chen, and M\. Deng\(2022\)ScNAME: neighborhood contrastive clustering with ancillary mask estimation for scrna\-seq data\.Bioinformatics38\(6\),pp\. 1575–1583\.Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p2.1)\.
- \[55\]J\. Wang, W\. Jiang, J\. Guan, and S\. Zhou\(2024\)Circle: scrna\-seq data clustering by cluster\-aware iterative contrastive learning\.In2024 IEEE International Conference on Bioinformatics and Biomedicine \(BIBM\),pp\. 1219–1225\.Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p2.1),[§III\-A](https://arxiv.org/html/2606.28459#S3.SS1.p1.1)\.
- \[56\]J\. Wang, A\. Ma, Y\. Chang, J\. Gong, Y\. Jiang, R\. Qi, C\. Wang, H\. Fu, Q\. Ma, and D\. Xu\(2021\)ScGNN is a novel graph neural network framework for single\-cell rna\-seq analyses\.Nature communications12\(1\),pp\. 1882\.Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p2.1),[§III\-A](https://arxiv.org/html/2606.28459#S3.SS1.p1.1)\.
- \[57\]L\. Wang, K\. Liu, and Y\. Yuan\(2025\-10\)GT\-a2t: graph tensor alliance attention network\.IEEE/CAA Journal of Automatica Sinica12\(10\),pp\. 2165–2167\.External Links:ISSN 2329\-9274,[Link](http://dx.doi.org/10.1109/jas.2024.124863),[Document](https://dx.doi.org/10.1109/jas.2024.124863)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p5.1)\.
- \[58\]L\. Wang, Y\. Yuan, and X\. Luo\(2026\-02\)Advanced high\-order graph convolutional networks with assorted time\-frequency transforms\.IEEE/CAA Journal of Automatica Sinica13\(2\),pp\. 394–408\.External Links:ISSN 2329\-9274,[Link](http://dx.doi.org/10.1109/jas.2025.125429),[Document](https://dx.doi.org/10.1109/jas.2025.125429)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p2.1)\.
- \[59\]L\. Wang, Y\. Yuan, and X\. Luo\(2026\-05\)Graph tensor convolutional network\.IEEE Transactions on Systems, Man, and Cybernetics: Systems56\(5\),pp\. 3008–3024\.External Links:ISSN 2168\-2232,[Link](http://dx.doi.org/10.1109/tsmc.2026.3655418),[Document](https://dx.doi.org/10.1109/tsmc.2026.3655418)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p2.1)\.
- \[60\]Q\. Wang, H\. Wu, and X\. Luo\(2025\-12\)A convolution bias\-incorporated nonnegative latent factorization of tensors model for accurate representation learning to dynamic directed graphs\.IEEE Transactions on Systems, Man, and Cybernetics: Systems55\(12\),pp\. 8902–8914\.External Links:ISSN 2168\-2232,[Link](http://dx.doi.org/10.1109/tsmc.2025.3611792),[Document](https://dx.doi.org/10.1109/tsmc.2025.3611792)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p4.1)\.
- \[61\]S\. Wang, Y\. Liu, H\. Zhang, and Z\. Liu\(2025\)ScE2EGAE: enhancing single\-cell rna\-seq data analysis through an end\-to\-end cell\-graph\-learnable graph autoencoder with differentiable edge sampling\.Biology direct20\(1\),pp\. 66\.Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p2.1)\.
- \[62\]Y\. Wang, Z\. Tang, H\. Huang, J\. Li, Z\. Wang, Y\. Yu, C\. Zhang, J\. Li, H\. Dai, F\. Wang,et al\.\(2018\)Pulmonary alveolar type i cell population consists of two distinct subtypes that differ in cell fate\.Proceedings of the National Academy of Sciences115\(10\),pp\. 2407–2412\.Cited by:[§III\-A](https://arxiv.org/html/2606.28459#S3.SS1.p1.1)\.
- \[63\]L\. Wei, L\. Jin, and X\. Luo\(2024\-06\)A robust coevolutionary neural\-based optimization algorithm for constrained nonconvex optimization\.IEEE Transactions on Neural Networks and Learning Systems35\(6\),pp\. 7778–7791\.External Links:ISSN 2162\-2388,[Link](http://dx.doi.org/10.1109/tnnls.2022.3220806),[Document](https://dx.doi.org/10.1109/tnnls.2022.3220806)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p3.1)\.
- \[64\]F\. A\. Wolf, P\. Angerer, and F\. J\. Theis\(2018\)SCANPY: large\-scale single\-cell gene expression data analysis\.Genome biology19\(1\),pp\. 15\.Cited by:[§II\-A](https://arxiv.org/html/2606.28459#S2.SS1.p1.5)\.
- \[65\]D\. Wu, Y\. Hu, K\. Liu, J\. Li, X\. Wang, S\. Deng, N\. Zheng, and X\. Luo\(2025\-04\)An outlier\-resilient autoencoder for representing high\-dimensional and incomplete data\.IEEE Transactions on Emerging Topics in Computational Intelligence9\(2\),pp\. 1379–1391\.External Links:ISSN 2471\-285X,[Link](http://dx.doi.org/10.1109/tetci.2024.3437370),[Document](https://dx.doi.org/10.1109/tetci.2024.3437370)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p3.1)\.
- \[66\]D\. Wu, S\. Li, Y\. He, X\. Luo, and X\. Gao\(2026\-05\)Non\-gradient hash factor learning for high\-dimensional and incomplete data representation learning\.IEEE Transactions on Pattern Analysis and Machine Intelligence48\(5\),pp\. 5811–5826\.External Links:ISSN 1939\-3539,[Link](http://dx.doi.org/10.1109/tpami.2026.3653780),[Document](https://dx.doi.org/10.1109/tpami.2026.3653780)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p3.1)\.
- \[67\]D\. Wu, Z\. Li, Z\. Yu, Y\. He, and X\. Luo\(2025\-02\)Robust low\-rank latent feature analysis for spatiotemporal signal recovery\.IEEE Transactions on Neural Networks and Learning Systems36\(2\),pp\. 2829–2842\.External Links:ISSN 2162\-2388,[Link](http://dx.doi.org/10.1109/tnnls.2023.3339786),[Document](https://dx.doi.org/10.1109/tnnls.2023.3339786)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p4.1)\.
- \[68\]D\. Wu, C\. Liang, Y\. He, Y\. Qiao, and X\. Luo\(2026\-03\)Multimetric autoencoder for representing high\-dimensional and incomplete data\.IEEE Transactions on Systems, Man, and Cybernetics: Systems56\(3\),pp\. 1533–1546\.External Links:ISSN 2168\-2232,[Link](http://dx.doi.org/10.1109/tsmc.2025.3646863),[Document](https://dx.doi.org/10.1109/tsmc.2025.3646863)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p3.1)\.
- \[69\]D\. Wu, X\. Luo, Y\. He, and M\. Zhou\(2024\-03\)A prediction\-sampling\-based multilayer\-structured latent factor model for accurate representation to high\-dimensional and sparse data\.IEEE Transactions on Neural Networks and Learning Systems35\(3\),pp\. 3845–3858\.External Links:ISSN 2162\-2388,[Link](http://dx.doi.org/10.1109/tnnls.2022.3200009),[Document](https://dx.doi.org/10.1109/tnnls.2022.3200009)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p3.1)\.
- \[70\]D\. Wu, P\. Zhang, Y\. He, and X\. Luo\(2024\-03\)MMLF: multi\-metric latent feature analysis for high\-dimensional and incomplete data\.IEEE Transactions on Services Computing17\(2\),pp\. 575–588\.External Links:ISSN 2372\-0204,[Link](http://dx.doi.org/10.1109/tsc.2023.3331570),[Document](https://dx.doi.org/10.1109/tsc.2023.3331570)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p3.1)\.
- \[71\]H\. Wu, Y\. Qiao, and X\. Luo\(2024\-11\)A fine\-grained regularization scheme for non\-negative latent factorization of high\-dimensional and incomplete tensors\.IEEE Transactions on Services Computing17\(6\),pp\. 3006–3021\.External Links:ISSN 2372\-0204,[Link](http://dx.doi.org/10.1109/tsc.2024.3486171),[Document](https://dx.doi.org/10.1109/tsc.2024.3486171)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p4.1)\.
- \[72\]H\. Wu, Q\. Wang, X\. Luo, and Z\. Wang\(2025\-12\)Learning accurate representation to nonstandard tensors via a mode\-aware tucker network\.IEEE Transactions on Knowledge and Data Engineering37\(12\),pp\. 7272–7285\.External Links:ISSN 2326\-3865,[Link](http://dx.doi.org/10.1109/tkde.2025.3617894),[Document](https://dx.doi.org/10.1109/tkde.2025.3617894)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p4.1)\.
- \[73\]M\. Wu, P\. Hu, Z\. You, J\. Zhang, L\. Hu, and X\. Luo\(2026\-06\)Graph\-based prediction of mirna\-drug associations with multisource information and metapath enhancement matrices\.IEEE Journal of Biomedical and Health Informatics30\(6\),pp\. 4513–4524\.External Links:ISSN 2168\-2208,[Link](http://dx.doi.org/10.1109/jbhi.2025.3558303),[Document](https://dx.doi.org/10.1109/jbhi.2025.3558303)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p2.1)\.
- \[74\]P\. Xu, Z\. Ning, M\. Xiao, G\. Feng, X\. Li, Y\. Zhou, and P\. Wang\(2024\)ScCDCG: efficient deep structural clustering for single\-cell rna\-seq via deep cut\-informed graph embedding\.InInternational Conference on Database Systems for Advanced Applications,pp\. 172–187\.Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p2.1),[§III\-A](https://arxiv.org/html/2606.28459#S3.SS1.p1.1)\.
- \[75\]P\. Xu, Z\. Wang, Z\. Wang, P\. Li, J\. Wang, R\. Zhang, P\. Wang, and Y\. Zhou\(2026\)ScCluBench: comprehensive benchmarking of clustering algorithms for single\-cell rna sequencing\.InProceedings of the AAAI Conference on Artificial Intelligence,Vol\.40,pp\. 1364–1372\.Cited by:[§III\-C](https://arxiv.org/html/2606.28459#S3.SS3.p1.1)\.
- \[76\]X\. Xu, M\. Lin, X\. Luo, and Z\. Xu\(2025\-03\)An adaptively bias\-extended non\-negative latent factorization of tensors model for accurately representing the dynamic qos data\.IEEE Transactions on Services Computing18\(2\),pp\. 603–617\.External Links:ISSN 2372\-0204,[Link](http://dx.doi.org/10.1109/tsc.2025.3544123),[Document](https://dx.doi.org/10.1109/tsc.2025.3544123)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p4.1)\.
- \[77\]X\. Xu, M\. Lin, Z\. Xu, and X\. Luo\(2025\-04\)Attention\-mechanism\-based neural latent\-factorization\-of\-tensors model\.ACM Transactions on Knowledge Discovery from Data19\(4\),pp\. 1–27\.External Links:ISSN 1556\-472X,[Link](http://dx.doi.org/10.1145/3719295),[Document](https://dx.doi.org/10.1145/3719295)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p4.1)\.
- \[78\]X\. Xu, M\. Lin, Z\. Xu, and X\. Luo\(2026\)A sampling\-neighborhood\-regularized latent factorization of tensor for dynamic qos estimation\.IEEE Transactions on Network and Service Management23,pp\. 1707–1722\.External Links:ISSN 2373\-7379,[Link](http://dx.doi.org/10.1109/tnsm.2025.3644937),[Document](https://dx.doi.org/10.1109/tnsm.2025.3644937)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p4.1)\.
- \[79\]H\. Yang, M\. Lin, H\. Chen, X\. Luo, and Z\. Xu\(2025\-01\)Latent factor analysis model with temporal regularized constraint for road traffic data imputation\.IEEE Transactions on Intelligent Transportation Systems26\(1\),pp\. 724–741\.External Links:ISSN 1558\-0016,[Link](http://dx.doi.org/10.1109/tits.2024.3486529),[Document](https://dx.doi.org/10.1109/tits.2024.3486529)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p4.1)\.
- \[80\]Y\. Yang, L\. Hu, G\. Li, D\. Li, P\. Hu, and X\. Luo\(2025\-09\)FMvPCI: a multiview fusion neural network for identifying protein complex via fuzzy clustering\.IEEE Transactions on Systems, Man, and Cybernetics: Systems55\(9\),pp\. 6189–6202\.External Links:ISSN 2168\-2232,[Link](http://dx.doi.org/10.1109/tsmc.2025.3578348),[Document](https://dx.doi.org/10.1109/tsmc.2025.3578348)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p1.1)\.
- \[81\]Y\. Yang, L\. Hu, G\. Li, D\. Li, P\. Hu, and X\. Luo\(2025\-08\)Link\-based attributed graph clustering via approximate generative bayesian learning\.IEEE Transactions on Systems, Man, and Cybernetics: Systems55\(8\),pp\. 5730–5743\.External Links:ISSN 2168\-2232,[Link](http://dx.doi.org/10.1109/tsmc.2025.3572738),[Document](https://dx.doi.org/10.1109/tsmc.2025.3572738)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p2.1)\.
- \[82\]M\. D\. Young, T\. J\. Mitchell, F\. A\. Vieira Braga, M\. G\. Tran, B\. J\. Stewart, J\. R\. Ferdinand, G\. Collord, R\. A\. Botting, D\. Popescu, K\. W\. Loudon,et al\.\(2018\)Single\-cell transcriptomes from human kidneys reveal the cellular identity of renal tumors\.Science361\(6402\),pp\. 594–599\.Cited by:[§III\-A](https://arxiv.org/html/2606.28459#S3.SS1.p1.1)\.
- \[83\]X\. Yu, X\. Xu, J\. Zhang, and X\. Li\(2023\)Batch alignment of single\-cell transcriptomics data using deep metric learning\.Nature communications14\(1\),pp\. 960\.Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p2.1),[§III\-A](https://arxiv.org/html/2606.28459#S3.SS1.p1.1)\.
- \[84\]Z\. Yu, Y\. Lu, Y\. Wang, F\. Tang, K\. Wong, and X\. Li\(2022\)Zinb\-based graph embedding autoencoder for single\-cell rna\-seq interpretations\.InProceedings of the AAAI conference on artificial intelligence,Vol\.36,pp\. 4671–4679\.Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p2.1)\.
- \[85\]Z\. Yu, Y\. Su, Y\. Lu, Y\. Yang, F\. Wang, S\. Zhang, Y\. Chang, K\. Wong, and X\. Li\(2023\)Topological identification and interpretation for single\-cell gene regulation elucidation across multiple platforms using scmgca\.Nature Communications14\(1\),pp\. 400\.Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p2.1),[§III\-A](https://arxiv.org/html/2606.28459#S3.SS1.p1.1)\.
- \[86\]Y\. Yuan, Q\. He, X\. Luo, and M\. Shang\(2022\-06\)A multilayered\-and\-randomized latent factor model for high\-dimensional and sparse matrices\.IEEE Transactions on Big Data8\(3\),pp\. 784–794\.External Links:ISSN 2372\-2096,[Link](http://dx.doi.org/10.1109/tbdata.2020.2988778),[Document](https://dx.doi.org/10.1109/tbdata.2020.2988778)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p3.1)\.
- \[87\]Y\. Yuan, J\. Li, and X\. Luo\(2024\-07\)A fuzzy pid\-incorporated stochastic gradient descent algorithm for fast and accurate latent factor analysis\.IEEE Transactions on Fuzzy Systems32\(7\),pp\. 4049–4061\.External Links:ISSN 1941\-0034,[Link](http://dx.doi.org/10.1109/tfuzz.2024.3389733),[Document](https://dx.doi.org/10.1109/tfuzz.2024.3389733)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p3.1)\.
- \[88\]Y\. Yuan, S\. Lu, and X\. Luo\(2025\-06\)A proportional integral controller\-enhanced non\-negative latent factor analysis model\.IEEE/CAA Journal of Automatica Sinica12\(6\),pp\. 1246–1259\.External Links:ISSN 2329\-9274,[Link](http://dx.doi.org/10.1109/jas.2024.125055),[Document](https://dx.doi.org/10.1109/jas.2024.125055)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p3.1)\.
- \[89\]Y\. Yuan, X\. Luo, M\. Shang, and Z\. Wang\(2023\-09\)A kalman\-filter\-incorporated latent factor analysis model for temporally dynamic sparse data\.IEEE Transactions on Cybernetics53\(9\),pp\. 5788–5801\.External Links:ISSN 2168\-2275,[Link](http://dx.doi.org/10.1109/tcyb.2022.3185117),[Document](https://dx.doi.org/10.1109/tcyb.2022.3185117)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p4.1)\.
- \[90\]Y\. Yuan, X\. Luo, M\. Shang, and D\. Wu\(2020\-04\)A generalized and fast\-converging non\-negative latent factor model for predicting user preferences in recommender systems\.InProceedings of The Web Conference 2020,WWW ’20,pp\. 498–507\.External Links:[Link](http://dx.doi.org/10.1145/3366423.3380133),[Document](https://dx.doi.org/10.1145/3366423.3380133)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p3.1)\.
- \[91\]Y\. Yuan, X\. Luo, and M\. Zhou\(2024\-04\)Adaptive divergence\-based non\-negative latent factor analysis of high\-dimensional and incomplete matrices from industrial applications\.IEEE Transactions on Emerging Topics in Computational Intelligence8\(2\),pp\. 1209–1222\.External Links:ISSN 2471\-285X,[Link](http://dx.doi.org/10.1109/tetci.2023.3332550),[Document](https://dx.doi.org/10.1109/tetci.2023.3332550)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p3.1)\.
- \[92\]Y\. Yuan, M\. Shang, and X\. Luo\(2020\)Temporal web service qos prediction via kalman filter\-incorporated dynamic latent factor analysis\.InECAI 2020,External Links:ISSN 0922\-6389,[Link](http://dx.doi.org/10.3233/faia200139),[Document](https://dx.doi.org/10.3233/faia200139)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p4.1)\.
- \[93\]Y\. Yuan, R\. Wang, G\. Yuan, and L\. Xin\(2023\-10\)An adaptive divergence\-based non\-negative latent factor model\.IEEE Transactions on Systems, Man, and Cybernetics: Systems53\(10\),pp\. 6475–6487\.External Links:ISSN 2168\-2232,[Link](http://dx.doi.org/10.1109/tsmc.2023.3282950),[Document](https://dx.doi.org/10.1109/tsmc.2023.3282950)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p3.1)\.
- \[94\]Y\. Yuan, S\. Wang, H\. Zhou, L\. Wang, and X\. Luo\(2026\)A novel approach to temporal qos estimation via extended kalman filter\-incorporated latent feature analysis\.IEEE Transactions on Services Computing,pp\. 1–12\.External Links:ISSN 2372\-0204,[Link](http://dx.doi.org/10.1109/tsc.2026.3697552),[Document](https://dx.doi.org/10.1109/tsc.2026.3697552)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p4.1)\.
- \[95\]Y\. Yuan, Y\. Wang, and X\. Luo\(2025\-06\)A node\-collaboration\-informed graph convolutional network for highly accurate representation to undirected weighted graph\.IEEE Transactions on Neural Networks and Learning Systems36\(6\),pp\. 11507–11519\.External Links:ISSN 2162\-2388,[Link](http://dx.doi.org/10.1109/tnnls.2024.3514652),[Document](https://dx.doi.org/10.1109/tnnls.2024.3514652)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p2.1)\.
- \[96\]N\. Zeng, X\. Li, P\. Wu, H\. Li, and X\. Luo\(2024\-02\)A novel tensor decomposition\-based efficient detector for low\-altitude aerial objects with knowledge distillation scheme\.IEEE/CAA Journal of Automatica Sinica11\(2\),pp\. 487–501\.External Links:ISSN 2329\-9274,[Link](http://dx.doi.org/10.1109/jas.2023.124029),[Document](https://dx.doi.org/10.1109/jas.2023.124029)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p1.1)\.
- \[97\]G\. X\. Zheng, J\. M\. Terry, P\. Belgrader, P\. Ryvkin, Z\. W\. Bent, R\. Wilson, S\. B\. Ziraldo, T\. D\. Wheeler, G\. P\. McDermott, J\. Zhu,et al\.\(2017\)Massively parallel digital transcriptional profiling of single cells\.Nature communications8\(1\),pp\. 14049\.Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p1.1)\.
- \[98\]Y\. Zhong, K\. Liu, S\. Gao, and X\. Luo\(2024\-10\)Alternating\-direction\-method of multipliers\-based adaptive nonnegative latent factor analysis\.IEEE Transactions on Emerging Topics in Computational Intelligence8\(5\),pp\. 3544–3558\.External Links:ISSN 2471\-285X,[Link](http://dx.doi.org/10.1109/tetci.2024.3420735),[Document](https://dx.doi.org/10.1109/tetci.2024.3420735)Cited by:[§I](https://arxiv.org/html/2606.28459#S1.p3.1)\.