TransBind2: Improving Transcription Factor-DNA Binding Prediction with Multimodal Data and Bidirectional Cross Attention

Le résumé fourni par la source

Dataset for TransBind-2, a protein-aware deep learning model for transcription factor (TF) binding site prediction using DNase-seq accessibility, genome uniqueness/mappability, and ProstT5 protein embeddings with bidirectional cross-attention fusion. Dnase_data.zip — DNase-seq chromatin accessibility signal (HDF5): `train_dnase.h5`, `val_dnase.h5`, `test_dnase.h5`. train_val_test_data.zip — One-hot DNA sequences and binding labels/indices for train, validation, and test sets: `train_unique_dna.npz`, `standalone_train_indices.npz`, `val_unique_dna.npz`, `standalone_val_indices.npz`, `test_unique_dna.npz`, `standalone_test_indices.npz`, `standalone_test_indices_with_expidx.npz` uniqueness.zip — Genome uniqueness/mappability scores: `train_uniqueness.npy`, `val_uniqueness.npy`, `test_uniqueness.npy`. protein_features.zip — ProstT5 protein embeddings for each transcription factor (`prostt5_featuresV1/`, one `.fea` file per TF). tf_mapping.zip — Mapping between TF names, UniProt IDs, and their corresponding protein feature files (`tf_mapping_with_features_with_graphsV1.csv`). TransBind-2.ckpt.gz — Trained TransBind-2 model checkpoint Code for data preprocessing, model training, and evaluation is available at https://github.com/jianlin-cheng/TransBind2

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.