The molecular signatures of neurological disorders arise from the specific contribution of many cell types. Cell type-specific dissection of such signatures is critical to characterizing complex underlying mechanisms driving such disorders and to efficiently deriving therapeutic targets. Although single-nucleus technologies have led to key findings in mechanisms of brain diseases, the limited nuclear transcriptional coverage coupled with the high cost and advanced technical skills required have hindered their broad applicability and potential to unveil robust cell type-specific candidates. Here we develop and validate an AI-based framework that enables low-cost and large-scale cell type-specific investigation of comprehensive transcriptional programs from bulk RNA-seq, significantly outperforming previous methods. Our approach leverages the ability of the transformer model to digest a large volume of data and learn a generalizable mapping, thereby restoring cell type-specific RNA profile from bulk expression. Notably, it allows the investigation of cell type-specific transcriptomic programs in complex and heterogeneous phenotypes such as cognitive resilience or brain resistance to AD. In multiple regions of the cerebral cortex, we identify astrocytes as the major cell mediator of resilience. In contrast, excitatory neurons and oligodendrocyte progenitor cells are identified as the major cell mediators of resistance, with putative molecular candidates maintaining synaptic function and preserving neuron health. Finally, using an optimally preserved mouse brain, we show the potential of our approach to restore the whole tissue cell-type specific transcriptome, rather than single nucleus only, offering a framework for an unbiased investigation of whole brain cell transcriptome.