LA4SR purified algal protein sequences from TARA-Oceans, MMETSP, and OSD assemblies [greedy decoding]

Algal protein sequences extracted from TARA-Oceans metatranscriptome assemblies, MMETSP transcriptomes, and Ocean Sampling Day (OSD) metagenomes using Al Khidr (Pythia-LA4SR), a Pythia-70M-based language model (checkpoint 55000) fine-tuned to classify algal versus non-algal protein sequences via greedy decoding. Each FASTA file contains the predicted algal proteins from one assembly. These sequences complement the algaGPT-based extraction and are used for comparative validation of the dual-classifier pipeline described in the accompanying manuscript. The companion dataset extracted using the algaGPT classifier is available at DOI:10.5281/zenodo.18728837. Analysis tables and model outputs: DOI:10.5281/zenodo.18538439.