There is interest in the use of combinatorial New Approach Methodologies, by integrating computational, statistical and machine learning based methods with human complex in-vitro systems such as brain organoids, to complement research and translational studies in humans and animal models. We evaluated the use of machine learning based methods (multi-layer perceptron models) to integrate transcriptome-wide single-cell non-spatial and spatial transcriptomic signatures across herpesvirus-infected 2D and 3D cerebral organoids with human fetal, childhood and adult brains. These machine learning based models achieved higher validation rates compared to a majority voting best-match approach, across several single-cell RNA-sequence datasets from human brain samples. Using our approach, we identified key cell types that were dysregulated by herpes simplex virus 1 (HSV-1) from our 2D single-cell RNA-sequence data. We identified cell type specific spatial differences in nearest neighbor distances that were significantly reduced or increased from our 3D single-cell RNA-sequence data. Some of these spatial differences were no longer significant when we compared the distances to those calculated from random permutations, indicating that these spatial differences were due to global differences in cell type density due to virus infections. As a future direction, we will use our systems and tools to characterize therapeutic modalities to reverse neuroinflammation-induced transcriptomic perturbations in Alzheimer’s disease and related dementias.