The DeepNEU platform specification
The current DeepNEU database (v6.1) contains 4516 gene/protein or phenotypic concepts and 41493 nonzero relationships resulting in a large amount of information flowing into and out of each node in the fully connected recurrent network. On average, each node in the network initially has >9 inputs and >9 outputs. An updated analysis of all positive and negative network connections revealed a bias toward positive outputs. The pretest probability of a positive outcome prediction is 0.656 and the pretest probability of a negative prediction is therefore 0.344. This system bias was used when applying the binomial test to all simulation outcomes.
The aiWBO Wild Type Simulations
The unsupervised whole brain organoid (aiWBO) simulations converged quickly (45 iterations) to a new system wide steady state without evidence of overtraining after 1000 iterations.
Neural Cell Types
The aiWBO correctly predicted the expression of all 6 neural cell types commonly found in human brain organoids. The probability that all (N = 6) cell type outcomes were predicted by chance alone using the binomial test is 0.08. These results are summarized in Figure 1A.
Rostral-Caudal Brain Regions
The aiWBO correctly predicted the expression of all 3 regions commonly found in human brain organoids. These regions are Forebrain, Midbrain and Hindbrain. The spinal cord was not simulated in the current version. The probability that the expression of all (N = 3) regions were predicted by chance alone using the binomial test is 0.282. These results are summarized in Figure 1B.
Ventral-Dorsal Brain Regions
The aiWBO correctly predicted the expression of all 7 ventral (anterior)-dorsal (posterior) regions commonly found in human brain organoids. The probability that the expression of all (N = 7) regions were predicted by chance alone using the binomial test is 0.052. These results are summarized in Figure 1C.
Acid-Base status of the organoid
The aiWBO predicted the expression of all 7 concepts representative of a mixed (compensated) metabolic-alkalosis and respiratory acidosis in human brain organoids. The probability that the expression of all (N = 7) concepts associated with the expression of a compensated metabolic-alkalosis and respiratory acidosis were predicted by chance alone using the binomial test is 0.052. These results are summarized in Figure 1D.
Cerebral Cortical Layers
The aiWBO correctly predicted the expression of all 6 cerebral cortical layers commonly found in human brain organoids. The probability that the expression of all (N = 6) cerebral cortical layers were predicted by chance alone using the binomial test is 0.08. These results are summarized in Figure 2A.
Cerebellar Cortical Layers
The aiWBO correctly predicted the expression of all 4 cerebellar layers cortical commonly found in human brain organoids. The probability that the expression of all (N = 4) cerebral cortical layers were predicted by chance alone using the binomial test is 0.185. These results are summarized in Figure 2B.
Microcirculation
The aiWBO correctly predicted the expression of all 7 concepts representative of a microcirculation in human brain organoids. The probability that the expression of all (N = 7) concepts associated with the expression of a microcirculation were predicted by chance alone using the binomial test is 0.052. These results are summarized in Figure 2C.
Blood Brain Barrier (BBB)
The aiWBO correctly predicted the expression of all 6 concepts representative of a BBB in human brain organoids. The probability that the expression of all (N = 6) concepts associated with the expression of a BBB were predicted by chance alone using the binomial test is 0.185. These results are summarized in Figure 2D.
Summary Results for aiWBO simulations
Taken together the aiWBO correctly predicted the expression of 39 elements consistent with a pattern seen in a whole brain organoid and 7 elements consistent with the presence of a compensated metabolic-alkalosis and respiratory acidosis. The probability that the expression of all (N = 46) concepts were predicted by chance alone using the binomial test is <0.00000001.
The aiWBO-APOE4 Alzheimer’s Disease (AD) Simulations
The unsupervised whole brain organoid (aiWBO-APOE4) AD simulations converged quickly (42 iterations) to a new system wide steady state without evidence of overtraining after 1000 iterations
Neural Cell Types
The aiWBO-APOE4 (AD) simulations correctly predicted the expression of all 6 neural cell types commonly found in human brain organoids. The probability that all (N = 6) cell type outcomes were predicted by chance alone using the binomial test is 0.08. A statistical analysis using the Mann-Whitney u test indicated that there were significant decreases in the expression of astrocytes and oligodendrocytes with the AD simulations. All other cell types including neurons were not significantly different between aiWBO and AD simulations. These results are summarized in Figure 3A.
While the major cell target of advanced AD is neurons, deleterious effects on synapses are among the earliest pathologic changes. When we examined the impact of increased APOE4 on the AD simulations using a 7-element, literature validated profile, the effects were largely consistent with significant decline in synaptogenesis and synaptic function. These results are summarized in Table 6S.
Rostral-Caudal Brain Regions
The aiWBO-APOE4 (AD) simulations correctly predicted the expression of all 3 regions commonly found in human brain organoids. These regions are Forebrain, Midbrain and Hindbrain. The spinal cord was not simulated in the current version. The probability that the expression of all (N = 3) regions were predicted by chance alone using the binomial test is 0.282. A statistical analysis using the Mann-Whitney u test indicated that there was a small but significant (p<0.05) decrease in the expression of the Forebrain and small but significant increase in the Midbrain while the Hindbrain was unchanged in the AD simulations. These results are summarized in Figure 3B.
Ventral-Dorsal Brain Regions
The aiWBO-APOE4 (AD) simulations correctly predicted the expression of all 7 ventral (anterior)-dorsal (posterior) representative regions commonly found in human brain organoids. The probability that the expression of all (N = 7) regions were predicted by chance alone using the binomial test is 0.052. A statistical analysis using the Mann-Whitney u test indicated that there were no significant differences (p>0.05) in the expression of all Ventral-Dorsal brain regions in the aiWBO and AD simulations. These results are summarized in Figure 3C.
Cerebral Cortical Layers
The aiWBO-APOE4 (AD) simulations correctly predicted the expression of all 6 cerebral cortical layers commonly found in human brain organoids. The probability that the expression of all (N = 6) cerebral cortical layers were predicted by chance alone using the binomial test is 0.08. A statistical analysis using the Mann-Whitney u test indicated that there was a significant decrease (p<0.01) in the expression of the Layers 2-5 with a very small but still significant decrease (p<0.05) Layer 6. Layer 1 was unchanged in the AD simulations. Overall, there was also a significant decrease in cerebral cortical mass in AD simulations. These results are summarized in Figure 3D.
Cerebellar Cortical Layers
The aiWBO-APOE4 (AD) simulations correctly predicted the expression of all 4 cerebellar cortical layers commonly found in human brain organoids. The probability that the expression of all (N = 4) cerebral cortical layers were predicted by chance alone using the binomial test is 0.185. A statistical analysis using the Mann-Whitney u test indicated that there were no significant differences (p>0.05) between expression of all cerebellar cortical layers in the aiWBO and AD simulations. These results are summarized in Figure 3E.
Microcirculation
The aiWBO-APOE4 (AD) simulations correctly predicted the expression of all 7 concepts representative of a microcirculation in human brain organoids. The probability that the expression of all (N = 7) concepts associated with the expression of a microcirculation were predicted by chance alone using the binomial test is 0.052. A statistical analysis using the Mann-Whitney u test indicated that there were significant decreases (p<0.05) in the endothelial-arterial, venous and lymphatic components of the Microcirculation in the AD simulations. In addition, the intracellular O2 concentration is significantly decreased (p<0.05) in the AD simulations consistent with a degree of microcirculation impairment. These results are summarized in Figure4A.
Blood Brain Barrier (BBB)
The aiWBO-APOE4 (AD) correctly predicted the expression of all 6 concepts representative of a BBB in human brain organoids. The probability that the expression of all (N = 6) concepts associated with the expression of a BBB were predicted by chance alone using the binomial test is 0.185. A statistical analysis using the Mann-Whitney u test indicated that there were significant decreases in the astrocyte (p<0.05) and pericyte (p<0.01) components of the BBB. In addition, the function of the BBB was significantly decreased (p<0.01) in the AD simulations while the expression of the BBB itself was not different. These results are summarized in Figure 4B.
Acid-Base status of the organoid
The aiWBO-APOE4 (AD) simulations predicted the expression of all 7 concepts representative of a mixed (compensated) metabolic-alkalosis and respiratory acidosis in human brain organoids. The probability that the expression of all (N = 7) concepts associated with the expression of a compensated metabolic-alkalosis and respiratory acidosis were predicted by chance alone using the binomial test is 0.052. A statistical analysis using the Mann-Whitney u test indicated that there were significant changes (p<0.05) in all components of the AD simulations compared with aiWBO. Most notable there is a significant increase in intracellular H+ concentration. Taken together these data suggest a somewhat less well compensated mixed metabolic alkalosis- respiratory acidosis in the AD simulations. The persistent high CO2 concentration is also consistent with the increased CO2 concentration (5%) that is part of the cocktail used in generating aiWBO and AD simulations. These results are summarized in Figure 4C.
The aiWBO-APOE4 Alzheimer’s Disease (AD) simulations
The aiWBO-APOE4 simulations predicted the expression of all 10 concepts, representative of the Alzheimer’s Disease (AD) disease profile, in a simulated human brain organoid. The probability that the expression of all (N = 10) AD concepts were predicted by chance alone using the binomial test is 0.015. A statistical analysis using the Mann-Whitney u test indicated that there were significant changes (p<0.001) in all components of the simulated AD disease profile, except for the expression of amyloid precursor protein (APP) when compared with aiWBO. These results are summarized in Figure 4D.
Summary of AD simulations
While the AD simulations produced some significant variability when compared with the aiWBO simulation outputs, the AD simulations also correctly predicted the expression of 39 elements consistent with a pattern seen in a whole brain organoid and 7 elements suggesting the expression of a compensated metabolic alkalosis and respiratory acidosis. The probability that the expression of all (N = 46) concepts were predicted by chance alone using the binomial test is <0.00000001.