Patient-derived neural cultures can reveal cellular abnormalities that are obscured by symptom-based psychiatric categories, yet drug-response studies commonly reduce a changing biological process to one observation time and one readout. This study asks whether the shape of a neuroimmune perturbation response can distinguish schizophrenia, major depressive disorder, and bipolar disorder while identifying the pharmacological exposure that produces the broadest cellular recovery. A biologically constrained simulation generated 120 virtual donor lines, equally divided among the three disorders and unaffected controls. Each donor contributed triplicate cortical-organoid observations under an unchallenged condition, tumor necrosis factor alpha (TNF-\(\alpha\)) alone, or TNF-\(\alpha\) followed by loxapine, ketamine, or lithium. Network synchrony, neurite complexity, mitochondrial membrane potential, excitation–inhibition balance, and interleukin-6 secretion were sampled at 0, 6, 24, and 72~h, yielding 7,200 records and 36,000 outcome values. Trajectory-encoded perturbation phenotyping represented every donor by interval changes and curvature terms rather than isolated values. Five-fold grouped cross-validation produced 0.700 balanced accuracy for four-class donor classification, with class-wise F1 values of 0.900, 0.585, 0.632, and 0.690 for control, schizophrenia, major depression, and bipolar disorder, respectively. Integrated recovery fractions showed disorder-selective maxima: loxapine in schizophrenia (0.506, 95% bootstrap interval 0.467–0.547), ketamine in major depression (0.499, 0.471–0.528), and lithium in bipolar disorder (0.571, 0.538–0.604). Removing interleukin-6 features reduced balanced accuracy most strongly, from 0.700 to 0.650, whereas removing excitation–inhibition features increased it to 0.758, indicating partial redundancy. These simulation findings answer the study question affirmatively within the stated generative assumptions: temporal, multireadout responses contain diagnostic and treatment-ranking information that static measurements discard. The results define falsifiable expectations for a donor-blocked organoid study and do not constitute clinical evidence.
Patient-derived neural cultures can reveal cellular abnormalities that are obscured by symptom-based psychiatric categories, yet drug-response studies commonly reduce a changing biological process to one observation time and one readout. This study asks whether the shape of a neuroimmune perturbation response can distinguish schizophrenia, major depressive disorder, and bipolar disorder while identifying the pharmacological exposure that produces the broadest cellular recovery. A biologically constrained simulation generated 120 virtual donor lines, equally divided among the three disorders and unaffected controls. Each donor contributed triplicate cortical-organoid observations under an unchallenged condition, tumor necrosis factor alpha (TNF-\(\alpha\)) alone, or TNF-\(\alpha\) followed by loxapine, ketamine, or lithium. Network synchrony, neurite complexity, mitochondrial membrane potential, excitation–inhibition balance, and interleukin-6 secretion were sampled at 0, 6, 24, and 72~h, yielding 7,200 records and 36,000 outcome values. Trajectory-encoded perturbation phenotyping represented every donor by interval changes and curvature terms rather than isolated values. Five-fold grouped cross-validation produced 0.700 balanced accuracy for four-class donor classification, with class-wise F1 values of 0.900, 0.585, 0.632, and 0.690 for control, schizophrenia, major depression, and bipolar disorder, respectively. Integrated recovery fractions showed disorder-selective maxima: loxapine in schizophrenia (0.506, 95% bootstrap interval 0.467–0.547), ketamine in major depression (0.499, 0.471–0.528), and lithium in bipolar disorder (0.571, 0.538–0.604). Removing interleukin-6 features reduced balanced accuracy most strongly, from 0.700 to 0.650, whereas removing excitation–inhibition features increased it to 0.758, indicating partial redundancy. These simulation findings answer the study question affirmatively within the stated generative assumptions: temporal, multireadout responses contain diagnostic and treatment-ranking information that static measurements discard. The results define falsifiable expectations for a donor-blocked organoid study and do not constitute clinical evidence.
