Pairing Heavy and Light Chains of Antibodies using Cross Attention

Note to user The files have been organized according to the directory structure that is needed for running the scripts. This dataset contains the following files : 1. PHeLiX-main/ PHeLiX GitHub repository files. 2. PHeLiX-main/dataset/ Preprocessed_datasets/Cleaned, filtered, and standardized versions of the original OAS and PairedAbNGS datasets used for downstream analyses. Combined_Data/FASTA file of all unique VH–VL sequence pairs obtained by merging the OAS and PairedAbNGS datasets. VHVL_bindings/Pickle (.pkl) dictionary mapping each VH sequence to the list of experimentally observed paired VL sequences. The VH and VL sequences correspond to the unique VH-VL pairs above ( Combined_Data/ ) VH_VL_Separated/FASTA files (uniq_onlyVH.fasta and uniq_onlyVL.fasta) of unique VH and VL sequences extracted from the combined dataset. onlyVH_clustering_75/CD-HIT clustering results for VH sequences at 75% sequence identity. onlyVL_clustering_75/CD-HIT clustering results for VL sequences at 75% sequence identity. graph_at_75/NetworkX graph object formed using VH and VL clustering at 75% sequence identity. This graph contains only the positive VH-VL edges. MODEL_OOD_SPLIT/FASTA files defining the Train/Validation/In-Distribution (MODEL) and Out-of-Distribution (OOD) splits for VH sequences, VL sequences, and paired VH–VL sequences. OOD_onlyVH_clustering_90/CD-HIT clustering results for OOD VH sequences at 90% sequence identity. OOD_onlyVL_clustering_90/CD-HIT clustering results for OOD VL sequences at 90% sequence identity. MODEL_onlyVH_clustering_90/CD-HIT clustering results for MODEL VH sequences at 90% sequence identity. MODEL_onlyVL_clustering_90/CD-HIT clustering results for MODEL VL sequences at 90% sequence identity. graph_based_negatives_OOD/Graph-based positive and negative pair generation outputs for the OOD dataset: STATS.pkl: NetworkX graph object corresponding to the OOD set, along with other graph statistics like node degree, unique edge counts, etc. MISSING_0.9.pkl: Missing (negative) graph edges. POSITIVES_0.9.fasta: Positive VH–VL pairs. NEGATIVES_0.9.fasta: Graph-sampled negative VH–VL pairs. graph_based_negatives_MODEL/Contains graph-based positive and negative pair generation outputs for the MODEL Train/Validation/In-Distribution dataset: STATS.pkl: NetworkX graph object corresponding to the MODEL set, along with other graph statistics like node degree, unique edge counts, etc. MISSING_0.9.pkl: Missing (negative) graph edges. POSITIVES_0.9.fasta: Positive VH–VL pairs. NEGATIVES_0.9.fasta: Graph-sampled negative VH–VL pairs. Germline_info/Pickle (.pkl) files with germline annotations for VH–VL cluster-based graph formed for MODEL set, at 90% sequence identity threshold. 3. External_Benchmarking_Datasets/ MAGE_RBD/ VH-VL pairs generated by MAGE, including the 969 filtered pairs and 20 selected pairs for experimental validation. SynAbLib/ Cleaned SynAbLib dataset, with 1,198,151 VH-VL sequence pairs.