Facial and Emotional Recognition Systems are technologies that primarily use AI and machine learning to analyze various inputs like facial expression, speech, and physiological signals, to identify an Show more
Facial and Emotional Recognition Systems are technologies that primarily use AI and machine learning to analyze various inputs like facial expression, speech, and physiological signals, to identify and classify human emotions and link them to a variety of epigenomic traits and states. We conducted a Meta-Meta Analysis via Pharmacogenomics (PGx) and Genome-Wide Association Studies (GWAS) across two separate manifestations, including facial physics and emotional expressions. Applying GWAS datasets, 10 GWAS datasets were included, and following multiple filtrations, a GWAS Meta-Meta analysis led to a Secondary Gene List (SGL) of 586 members. Additionally, various indepth silico analyses, such as Protein-Protein Interactions (PPIs), refined 300 genes into a unified network, then, by adding 10 GARS genes, 309 genes remained. A different analysis of PPIs uncovered 141 connected genes (Final Gene List: FGL); more precisely, we conducted a PGx-based approach on this FGL. Finally, 1,480 annotations were found, among them, 682 annotations were significant; thus, we considered the genes with at least one significant annotation and found 54 Pharmacogenes in FGL (PGx-FGL). Through this in-depth analysis, we identified strong, significant top phenotypic roles for both DRD2 and BDNF linking genes in 48,780,906 subjects. Our PGx-based GWAS meta-meta-analyses, coupled with genetic and epigenetic liability testing, connected Facial and Emotional Recognition Systems to Spectrum Disorders (Attention-Deficit Hyperactivity Disorder: ADHD and Autism), Schizophrenia, Depression, and Anxiety. We propose that these findings could have heuristic therapeutic targeting potential and, as such, require intensive further clinical support. Show less
Alzheimer's Disease (AD) drug discovery has been hampered by patient heterogeneity, and the lack of sensitive tools for precise stratification. Here, we demonstrate that our robust and interpretable A Show more
Alzheimer's Disease (AD) drug discovery has been hampered by patient heterogeneity, and the lack of sensitive tools for precise stratification. Here, we demonstrate that our robust and interpretable AI-guided tool (predictive prognostic model, PPM) enhances precision in patient stratification, improving outcomes and decreasing sample size for a AD clinical trial. The AMARANTH trial of lanabecestat, a BACE1 inhibitor, was deemed futile, as treatment did not change cognitive outcomes, despite reducing β-amyloid. Employing the PPM, we re-stratify patients precisely using baseline data and demonstrate significant treatment effects; that is, 46% slowing of cognitive decline for slow progressive patients at earlier stages of neurodegeneration. In contrast, rapid progressive patients did not show significant change in cognitive outcomes. Our results provide evidence for AI-guided patient stratification that is more precise than standard patient selection approaches (e.g. β-amyloid positivity) and has strong potential to enhance efficiency and efficacy of future AD trials. Show less