A Curated Phenotype-Linked HIV-1 Sequence Dataset for Machine Learning Approaches to Tropism Prediction, Disease Progression and Proviral Reservoir Studies

HIV-1 tropism toward the CCR5 and CXCR4 coreceptors is an important viral characteristic associated with cell entry, the potential use of specific entry inhibitors, studies of disease progression, and analyses of proviral reservoirs. The development of tools for genotypic tropism prediction also requires annotated sequence datasets with reliable phenotypic annotation. In this study, we describe a curated HIV-1 sequence dataset intended to complement, rather than replace, existing reference collections. Publications and GenBank records were searched manually using PubMed, Google Scholar, OpenAlex, cross-references, and records from the GenBank Nucleotide database. Sequences were included if a GenBank accession number could be linked to the V3 region of the HIV-1 env gene, or to a longer fragment containing this region, and to phenotypically determined tropism; records already present in LANL were removed. The final dataset comprises 1809 records associated with 32 unique PubMed-indexed publications. Subtype B predominated in the dataset; however, non-B variants accounted for 455 records (25.2%), including several Circulating Recombinant Forms (CRFs) and Unique Recombinant Forms (URFs). R5 variants accounted for 1528 records (84.5%), whereas X4 and R5/X4 variants accounted for 281 records (15.5%). The dataset includes 14 categories of phenotypic assays, geographic metadata, and sequences ranging from env gene fragments to near-complete HIV-1 genomes. This resource can be used as a complement to existing databases for the development, testing, and external validation of approaches for HIV-1 tropism prediction.