Public sequence databases reveal annotation heterogeneity and deposition bias in Cryptosporidium and Giardia molecular surveillance

Molecular subtyping of Cryptosporidium and Giardia duodenalis relies heavily on public sequence repositories, but annotation heterogeneity and uneven deposition may bias epidemiological interpretation. We audited GenBank and UniProtKB records for Cryptosporidium GP60 and G. duodenalis gdh, tpi and bg, with SRA and CryptoDB used for contextual checks. GenBank metadata were extracted using qualifier-anchored parsing, host strings were normalized into 17 categories, host structure was compared across loci, and temporal host-category richness was assessed by rarefaction. UniProt annotations were evaluated for domain completeness, taxonomic consistency and GP60 domain boundaries. Among 22,980 GenBank records, 18,405 of 18,887 host-bearing records (97.45%) were classified. C. hominis GP60 was predominantly human-associated (89.8%), whereas C. parvum showed a broader zoonotic profile (54.0% cattle; 30.6% human; chi-square=4440.7, df=64, p<0.0001). UniProt domain annotation completeness ranged from 0% for tpi and bg to 89.6% for GP60 and 100% for gdh; GP60 Pfam starts occurred at 50 positions. Rarefied host-category richness increased from 3.0 before 2005 to 12.8 in 2020+, showing that temporal structure in deposited records persisted after equalizing sampling effort. A common Entrez Direct extraction idiom exposed 83.7% of records to potential metadata column misalignment. These findings show that database curation and deposition intensity are measurable sources of uncertainty in Cryptosporidium/Giardia molecular surveillance. The corrected extraction workflow and host-classification framework provide reusable tools for more reproducible host-ecological and zoonotic inference from public sequence data.