Table 4.

The summary of computational methods/algorithms for EV analysis.

Methods/ Algorithms	Strengths	Weaknesses	Applications in EV analysis	Reference
PCA	Preserve original data and maximize eigenvalues	Challenging to analyze minor variations	EV transcriptomics/proteomics	[273, 277]
LDA	Optimize class distinction and surpass PCA in classification	Errors and overfitting risk with small sample size	Exosomal multi‐miRNA analysis	[278]
KNN/SVM	Effectiveness in high dimensional spaces and simplicity	Inefficient with missing data and costly for large datasets	Correlating urinary EV biomarkers signals with clinical states	[279]
XGBoost	Efficiency and scalability	Model complexity and time‐consuming	Identification of EVs from different sources	[280]
PLS‐DA	Effectiveness with small sample sizes	Model complexity and overfitting risk	Identifying candidate biomarkers in EV metabolomics	[274, 281]
PCA‐LDA	Improved classification accuracy and overcoming overfitting	Sensitivity to data scaling	Distinguishing cancerous from normal exosomes subtype	[282]
ResNet	Boosts training stability and generalization accuracy	Require more computational resources and optimization difficulty	Predicting lung cancer from cell and plasma exosomes correlation	[263, 283]
ANN	Effectively handle large datasets and nonlinear data	Require large amounts of data resources	Classifying breast cancer subtypes via EV SERS spectra	[264, 284]
CNN	High specificity and sensitivity	Requires substantial computational resources	Diagnosing MDD with Plasma EV Raman features	[269, 284]