ApexOracle paper assets: embeddings, MIC scoring, and result-reproduction data
收藏资源简介:
This version extends the ApexOracle paper-asset series while retaining the original precomputed genome and strain-text embedding archives. Genome_embs.tar.gz: precomputed Evo-2 genome embeddings. Text_Description.tar.gz: strain descriptions and precomputed text embeddings. mic_candidate_scorer_all_peptide_non_pad_t1e-3_epoch13.pth: the all-peptide post-generation MIC candidate-scoring checkpoint. apexoracle_fig1b_classification_reproduction_v1.tar.gz: exact Fig. 1b folds, normalized predictions, and a metric checker. apexoracle_fixed_mic_reconstruction_v1.tar.gz: 21-member post-paper fixed-split strain-wise MIC reconstruction, normalized sample-level predictions, exact reconstruction membership, and an independent metric checker. The fixed MIC reconstruction is explicitly not presented as the unrecovered 2025 checkpoint membership. The MIC candidate scorer uses the fixed-epsilon non-padding-preserved profile at t=1e-3 and is not the Generation sampler checkpoint. Machine-readable file roles, hashes, and scientific boundaries are provided in the version release manifest.



