This paper presents a full transparency audit and corrected analysis of the original GIIdrug paper (v1, DOI: 10. 5281/zenodo. 20115927). The v1 paper reported rho=-0. 971 between GIIdrug and skin irAE in n=8 ICI drugs, raising anti-circularity concerns. Audit findings: The v1 GII values are a monotone function of CDR3 heavy-chain loop length (CDR3H, SAbDab database), with GII vs CDR3H rho=1. 000 for the original 8 drugs. CDR3H is independently accessible from SAbDab before any clinical immunogenicity data is collected. Corrected claim: CDR3H (SAbDab, independent of clinical data) significantly predicts ADA rate (rho=0. 737, p=0. 0005) and skin irAE (rho=0. 821, p<0. 0001) across n=18 biologics spanning 7 target classes (PD-1, PD-L1, CTLA-4, TNF, HER2, VEGF, IL-6R). Molecular weight is non-predictive (rho=-0. 096, p=0. 70). Permutation tests confirm significance (N=10, 000, perm-p<0. 002). GIIdrug operationalizes CDR3H geometric complexity within the Lo-Shu framework.
Yao-Kai Kao (Tue,) studied this question.