Supplemental data for 'Vector leaf trees and satellite vision transformer embeddings decode macroalgal genome-environment couplings'

Marine macroalgae span extreme environments, but the genomic basis of their salinity and temperature tolerance is unclear. We sequenced nine Arabian Gulf macroalgae and combined them with 117 published genomes (126 total; 69 Rhodophyta, 43 Ochrophyta, 14 Chlorophyta) to test genome–environment links. Protein domain (Pfam) profiles were paired with two environmental frameworks: satellite-derived variables (sea surface temperature, salinity, chlorophyll-a, bathymetry via Google Earth Engine) and learned vision-transformer embeddings from petabyte-scale imagery (AlphaEarth Foundations). Gradient-boosted trees with phylum-aware cross-validation classified salinity at 87% ± 4.8% accuracy, indicating conserved osmotic solutions, whereas temperature prediction failed across phyla (R² ≈ 0) but improved within Ochrophyta (R² ≈ 0.28), pointing to lineage-restricted thermal architectures dominated by uncharacterized domains. Correlation screens (6,533 Pfams × 20 variables) yielded 76 FDR-significant and four Bonferroni-significant associations, led by DUF3570 (PF12094; r = -0.54 with temperature), and adhesion-related vWF-A domains were enriched in Arabian Gulf brown algae across both environmental frameworks. These results suggest salinity tolerance arises from conserved, multi-pathway mechanisms, while thermal tolerance mechanisms in macroalgae are lineage-specific and driven largely by unannotated protein families.