Table 2.

Machine learning methods to predict vaccine immunogenicity and efficacy. Different machine learning algorithms can be used. The quality of the model needs to be evaluated, and there are different metrics to assess a model performance, such as accuracy (defined as the number of correct predictions divided by the total number of input data), Area Under the Receiver Operator Characteristic curve (AUROC) or Root Mean Squared Error for regressions. It depends on the machine learning method itself. (Ab, antibody; ClaNC, classification to nearest centroid; DAMIP, discriminant analysis via mixed integer programming; HAI, hemagglutination-inhibition; CHMI, Controlled Human Malaria Infection; * accuracy except otherwise mentioned).

Vaccine	Vaccinees	Predicted Responses	Predictors	Machine Learning Method	Performance *	Reference
Yellow fever vaccine (YF-17D)	Healthy adults	The magnitude of the activated CD8+ T cell and neutralizing Ab responses	Early blood transcriptional signatures	ClaNC and DAMIP	Up to 90% and 100% respectively	[52]
Seasonal Trivalent Inactivated influenza Vaccine (TIV)	Patients 50–89 years old suffering from multiple chronic medical conditions	The magnitude of plasma HAI Ab response	Baseline signatures among 26 input continuous or categorical variables inc. previous vaccination, low grade chronic inflammation, chronic infections, blood cell counts	Neural network (multilayer perceptron (MLP), radial-basis function network (RBFN) and probabilistic network (PNN)) and Logistic regression	72.5% of average hit rate across 10 samples	[184]
Seasonal Trivalent Inactivated influenza Vaccine (TIV)	Healthy adults	The magnitude of plasma HAI Ab response	Early blood transcriptional signatures	DAMIP	Up to 90%	[185]
Seasonal Trivalent Inactivated influenza Vaccine (TIV)	Healthy adults, inc. young (20–30 years) and older subjects (60 to 89 years)	The magnitude of plasma HAI Ab response	Baseline blood transcriptional, cytokines and cell populations signatures	Logistic regression	84%	[178]
Seasonal Trivalent Inactivated influenza Vaccine (TIV) and pandemic H1N1 (pH1N1) vaccine	Healthy adults	The magnitude of the Ab response	Baseline HAI titer, blood cell populations, transcripts and pathways signatures	Diagonal linear discriminant analysis (for cell frequency data and when cell frequency and pathway status were combined); or partial least square (for data dimension reduction due to the large number of genes) followed by linear discriminant analysis (PLS-LDA) for transcript data alone	0.86 of AUROC	[60]
Seasonal Trivalent Inactivated influenza Vaccine (TIV) over 5 seasons	Human adults, inc. elderlies (>65 years)	The magnitude of plasma HAI Ab response	Early blood transcriptional signatures	DAMIP and artificial neural network classifier	>80%	[10]
Seasonal Trivalent Inactivated influenza Vaccine (TIV)	Healthy adults (50 to 74 years)	The magnitude of the B-cell ELISPOT and plasma HAI Ab responses	Early blood cell composition, mRNA-Seq, and DNA methylation signatures	The ensemble learner (inc. Generalized linear models, Recursive Partitioning, and Regression Trees), and random forest models	0.64–0.79 of AUROC	[186]
Seasonal Trivalent Inactivated influenza Vaccine (TIV)	Healthy adults	The magnitude of plasma HAI Ab response	Baseline HAI titer and blood transcriptional signatures	Gaussian Mixture Model (GMM)	R2 = 0.64 for the correlation between observed and predicted data	[187]
Seasonal Trivalent Inactivated influenza Vaccine (TIV)	Healthy adults	The magnitude of the Ab response	Early blood transcriptional signatures	Logistic Multiple Network-constrained Regression	69%	[188]
Seasonal Trivalent Inactivated influenza Vaccine (TIV) over 8 seasons	Healthy adults	The magnitude of the specific Ab response	Baseline blood cell populations signatures	128 machine learning algorithms suitable for classification using Sequential Iterative Modeling “OverNight” (SIMON), inc. Diagonal Discriminant Analysis, Partial Least Squares, Linear Discriminant Analysis, Logic Regression, Neural Network, Random Forest	Up to 0.92 of AUROC	[179]
Seasonal Trivalent Inactivated influenza Vaccine (TIV) given transcutaneously, intradermally or intramuscularly	Healthy adults	The magnitude of the specific T CD8+ and Ab responses	Early blood transcriptional and serum cytokines signatures	Logistic regression	0.93 to 0.96 of AUROC	[189]
Seasonal Trivalent Inactivated influenza Vaccine (TIV) and 23-valent pneumococcal polysaccharide vaccine	Old patients (>65 years) with chronic kidney disease with or without non-dialysis	The magnitude of the HAI Ab and anti-PnPS IgG responses	Baseline signatures among 30 input continuous or categorical variables inc. previous vaccinations, low grade chronic inflammation, chronic infections, blood cell counts	Multivariable linear regression model	p < 0.05	[190]
RTS,S malaria vaccine	Healthy adults	The protection against CHMI	Early blood transcriptional signatures	DAMIP	>80%	[181]
Candidate malaria vaccine composed of a Self-Assembling Protein Nanoparticles presenting the malarial circumsporozoite protein (CSP) adjuvanted with three different liposomal formulations: liposome plus Alum, liposome plus QS21, or both	Rhesus macaques	Adjuvant condition	Vaccine-induced immune response signatures among many variables inc. serology, fluorospot, ICS from blood, liver, LN and spleen	Random forest followed by Linear regression analysis	92%	[32]
Live-attenuated varicella zoster virus (VZV) vaccine	Healthy adults, inc. younger (25–40 years) and older (60–79 years)	The magnitude of the specific T and IgG responses	Early blood transcriptional, metabolite clusters, cytokines, and cell populations signatures	Multivariate regression model (Partial least square)	p < 0.05	[180]
Monovalent oral polio vaccine type 3 (mOPV3)	Infants aged 6–11 months	Seroconversion or shedding of vaccine virus as a marker of vaccine “take”	Baseline enteric pathogens blood cell populations, and plasma cytokines signatures	Random forest	58%	[191]
Two distinct live attenuated Tularemia vaccine administered by scarification	Healthy humans	The magnitude of the specific Ab and activated CD4 and CD8 T cell responses	Early blood transcriptional signatures	Logistic regression	26% of mean misclassification error	[39]
rVSV-ZEBOV	Healthy adults	The magnitude of the Ab response	Early blood transcriptional, plasma cytokine and cell populations signatures	Sparse partial least-squares followed by multivariable linear regression	0.77 of root square residuals leave-one-out explaining 55% of the variability	[12]
DNA/rAd5 HIV-1 preventive candidate vaccine	Healthy adults	HIV infection	Magnitude and quality of CD4 and CD8 T cells	PCA followed by Cox proportional hazards regression model, and Logistic regression with lasso	Up to 0.75 of AUROC	[192]
Seven preventive HIV-1 vaccine regimens (inc. DNA, NYVAC, ALVAC, MVA, AIDSVAX)	Healthy adults	The magnitude of long-term immune responses	Baseline demographic variables and peak immune responses	Regularized random forest and linear regression models	R = 0.91 for the correlation between observed andpredicted data	[193]
41 different vaccine vectors all expressing the same antigen	Mice	The quality of late T-cell responses	Early transcriptome of dendritic cells	Random forest	Up to 98%	[194]