Reprocessed data for training and benchmarking Deep Amino acid Selection Models

This consists of reprocessed data from various sources for training and benchmarking. This data is meant to be consumed by the dasm-experiments repository. Benchmark datasets: Koenig et al. (2017) - Anti-VEGF antibody deep mutational scanning Shanehsazzadeh et al. (2023) - Trastuzumab binding data from generative models Petersen et al. (2024) - MAGMA-seq data (2 antibody systems) Kirby et al. (2025) - MAGMA-seq data (4 antibody systems, SARS-CoV-2) Note: the first two data sets are from FLAb; data is from commit 67738ee (April 17, 2024). The FLAb repository has continued to receive updates; we use this specific commit for reproducibility. Training datasets: Tang et al. (2022) - DeepSHM heavy chain sequences Jaffe et al. (2022) - Paired heavy and light chain sequences Engelbrecht et al. (2025) - Heavy and light chain sequences (kappa and lambda) Rodriguez et al. (2023) - RACE-seq data for perplexity analysis