Systematic Comparison of Four TWAS Scaling Methods Within an Integrated Ki–DDD–TWAS Pipeline for Antipsychotic Metabolic Risk Assessment

Antipsychotic-associated metabolic abnormalities, including dyslipidemia, weight gain, hyperglycemia, and metabolic syndrome, vary substantially between drugs. We developed a reproducible computational pipeline that integrates receptor-binding affinity, defined daily dose, and precomputed transcriptome-wide association study data to produce drug-level metabolic-risk scores for antipsychotics. The pipeline retrieves drugs classified under ATC group N05A, normalizes defined daily doses to milligrams, fuzzy-matches drug names to a Ki database, aggregates receptor-binding measurements, maps receptors to genes, and integrates five lipid-related traits: low-density lipoprotein cholesterol, high-density lipoprotein cholesterol, log-transformed triglycerides, non-high-density lipoprotein cholesterol, and total cholesterol. Four transcriptome-wide association study scaling strategies were compared: a baseline without per-receptor scaling, linear scaling by absolute z score, exponential scaling by absolute z score, and exponential scaling by negative log10 p value. The final comparison included 52 drugs, five traits, and 1,040 drug–trait–method risk-score observations. Drug rankings were highly concordant across methods, with a mean Spearman correlation of 0.997. Only 38 of 260 drug–trait combinations showed a rank shift of at least three positions. Clozapine ranked first for every trait and method, while HRH1 and HTR2A were the two most consistent receptor-level contributors. The results indicate that the principal effect of changing the transcriptome-wide association study transformation was to alter score magnitude rather than drug prioritization. The negative-log10-p method is suitable as a primary analysis when evidence-weighted scaling is desired, while the baseline method is useful as a sensitivity analysis.Keywords: Antipsychotics; metabolic risk; transcriptome-wide association study; receptor pharmacology; Ki affinity; dose normalization; reproducible computing