Receptor Contribution Reallocation After Removal of Literature-Derived Weights: A Secondary Sensitivity Analysis of the Integrated Ki-DDD-TWAS Pipeline for Antipsychotic Lipid-Trait Prioritization

Background and objective Antipsychotic-associated metabolic abnormalities, including weight gain, dyslipidemia, hyperglycemia, obesity, and metabolic syndrome, vary among drugs. Receptor pharmacology offers mechanistic clues, while transcriptome-wide association studies identify genetically regulated expression signals linked to lipid traits. These evidence streams are usually evaluated separately. This secondary computational sensitivity analysis used previously generated Ki-DDD-TWAS outputs and was not a primary discovery or clinical validation study. The primary hypothesis was that removing literature-derived metabolic receptor weights would preserve overall drug ranking, with Spearman ρ approximately 0.9, while shifting receptor-level contributions from histaminergic and serotonergic systems toward dopaminergic and low-prior TWAS signals. Secondary objectives were to identify low-weight genes with strong TWAS signals, assess discordance with an author-defined clinical-liability comparator, and test sensitivity to gene removal and TWAS scaling. Methods Previously generated outputs were analyzed. Ki, the inhibition constant, and DDD, the defined daily dose, served as affinity and exposure-proxy inputs. Receptor-level and per-drug risk-score files were analyzed for 52 antipsychotics across LDL, HDL, log-transformed triglycerides, non-HDL, and total cholesterol. Analyses included reconstruction validation, pooled receptor contributions, TWAS signal versus literature weight, clinical-liability discordance, leave-one-gene-out and TWAS-scaling sensitivity, and receptor co-occurrence in high-risk drug sets. No patient-level data, independent TWAS resources, longitudinal outcomes, or outcome-based validation were used. Results Reconstruction reproduced saved risk ranks with Spearman correlation 1.000 for all 10 mode-trait combinations. V4 rankings remained concordant after removing literature weights, with trait-specific correlations of 0.900-0.923 and a mean of 0.913. The weighted model was dominated by HRH1, HTR2A, and HTR2C, contributing 25.7%, 21.6%, and 10.3% of pooled signal. Under uniform weighting, DRD2, DRD3, and HTR2A led, accounting for 22.1%, 12.9%, and 12.6%. ADRB1 was the strongest high-priority low-weight TWAS candidate, with an absolute HDL z-score of 13.53 and literature weight of 0.08. Other candidates included CHRM4, DRD2, DRD4, SLC6A4, and ADRB2. The model most strongly under-ranked paliperidone versus the supplied 12-drug comparator, while ziprasidone rose under uniform weighting. Removing HRH1 produced the largest rank shifts, particularly for levomepromazine, promazine, and acepromazine. Conclusions Removing literature-derived weights preserved overall drug ranking but substantially changed receptor-level score allocation. The weighted model emphasized histaminergic and serotonergic receptors prioritized in metabolic-liability literature, whereas the uniform model revealed broader dopaminergic and low-prior contributions. ADRB1 is a computational hypothesis-generating candidate, not a confirmed mechanism or established metabolic-risk contributor. HDL-association direction was not analyzed. The framework supports dual reporting of clinically anchored weighted and discovery-oriented uniform models as an internal sensitivity strategy only. It is not externally validated, does not predict patient-level outcomes, and should not be interpreted as estimating clinical metabolic risk or establishing receptor-level causality.