Data for "Entrenchment of germline amino-acid differences in antibody affinity maturation"

Data archive for reproducing the analyses in: Noam Harel, Kevin Sung, Will Dumm, Mackenzie M. Johnson, David Rich, Julia Fukuyama, Hugh K. Haddox, Frederick A. Matsen IV. "Entrenchment of germline amino-acid differences in antibody affinity maturation." bioRxiv (2026). https://doi.org/10.64898/2026.04.21.720000 This archive contains parent-child pair (PCP) datasets, trained deep learning model weights, and pre-computed intermediate results needed to reproduce the paper's figures and tables. Analysis code is available at https://github.com/matsengrp/dasm-epistasis-experiments. Datasets This study uses four human heavy-chain antibody repertoire datasets, previously processed by the DASM and Thrifty publications. Sequences from each dataset are partitioned into clonal families with inferred phylogenetic trees; parent-child pairs (PCPs) are extracted from sequential tree branches. Dataset Description Reference Jaffe Productive heavy-chain dataset. Jaffe et al. (2022) Tang Productive heavy-chain dataset. Tang et al. (2022); Vergani et al. (2017) Rodriguez Productive heavy-chain dataset. Rodriguez et al. (2023) Tang-SHM Out-of-frame (non-productive) dataset. Tang et al. (2022); Vergani et al. (2017) Naming convention: The Jaffe dataset files are prefixed wyatt-10x-1p5m in the archive. Archive contents v3/ — Productive PCP datasets and ANARCI numbering outputs: wyatt-10x-1p5m_fs-all-NoWinCheck_igh_pcp_2024-10-29_NI_noN_no-naive.csv.gz — Jaffe dataset tang-deepshm-prod-NoWinCheck_igh_pcp_2024-10-29_MASKED_NI_noN_no-naive.csv.gz — Tang dataset rodriguez-airr-seq-race-prod-NoWinCheck_igh_pcp_2024-11-12_MASKED_NI_noN_no-naive.csv.gz — Rodriguez dataset v3/anarci/ — Chothia-numbered sequence position mappings for each dataset v1/ — Out-of-frame PCP dataset: tang-deepshm-oof_pcp_2024-04-09_MASKED_NI.csv.gz — Tang-SHM out-of-frame dataset trained_models/ — DASM and DNSM model weights and metadata: dasm_4m-v1jaffeCC+v1tangCC-joint.pth — DASM model (4M parameters), trained on Jaffe + Tang dnsm_1m-v1jaffe+v1tang-joint.pth — DNSM model (1M parameters), trained on Jaffe + Tang Corresponding .yml config files and .csv branch length files for each model dasm_test_output/ — Cached DASM evaluation results on Rodriguez dnsm_test_output/ — Cached observed mutation datasets on Jaffe, Tang, and Rodriguez, computed via the DNSM framework for convenience neutral_mutability_cache/ — Pre-computed neutral mutability DataFrames for Jaffe, Tang, and Rodriguez The cached output directories (dasm_test_output, dnsm_test_output, neutral_mutability_cache) are included for convenience because they are computationally expensive to regenerate. The analysis code will recreate them automatically if they are missing. Only the PCP files, ANARCI outputs, and trained models are strictly required.