DS-Mamba Datasets for Protein Secondary Structure Prediction

It includes the Stage-1 PDB training dataset (PDB_Pure_Fixed.fasta), the Stage-2 training dataset (MMseqs_Cleaned_Trainset.fasta), and the independent 100-protein PlantOilCropTest dataset used for final evaluation. All protein records are stored in a three-line FASTA format containing a protein identifier, an amino acid sequence, and the corresponding residue-level eight-state secondary structure (Q8) labels.

The archive also provides the fixed seed-42 train/validation split files used in the reported experiments. After the training-time filtering implemented in the public DS-Mamba code, the Stage-1 dataset contains 6,858 training and 762 validation proteins, while the Stage-2 dataset contains 10,107 training and 1,124 validation proteins. PlantOilCropTest contains 100 independent plant-protein sequences and is used only for evaluation.