Molecular dynamics simulations of sequence-controlled Kremer-Grest copolymers with ten-monomer chains

Coarse-grained (Kremer-Grest bead-spring) molecular dynamics trajectories for 528 canonical sequence-controlled AB copolymers (10 monomers/chain, reduced from all 1024 binary sequences by collapsing 32 palindromic sequences and pairing the remaining 496 reverse-symmetric duplicates), 3 replicate seeds each. Includes KDE-rendered aggregate-morphology images (WebP-lossless), EfficientNet-B0 pooled feature embeddings, and PCA coordinates, used in the CDS&E "Generative Design of Copolymers" workshop (github.com/wfreinhart/copolymer-workshop-instructor).Two delivery tiers: core.tar.gz (~76 MB, the minimum data needed to run the workshop notebooks) and full_images.tar.gz (~85 MB, the complete 528x9 rendered-image library). Raw GSD trajectories (1584 files, 528 sequences times 3 seeds) ship as 16 randomly-shuffled shards (gsd_shard_01 through gsd_shard_16, ~54 MB each) so any single shard is representative of the full sequence space; sequence_to_shard.csv maps each sequence to its shard.Version 2 addition: random-walk-initialized (rwinit) arm. The original (v1) trajectories were initialized from a lattice of chains. This version adds a second, independent set of simulations covering the same 528 canonical sequences and 3 seeds, started instead from random-walk initial configurations. It exists as a control: some blocky-sequence morphologies in the lattice-initialized arm are artifacts of that initial configuration rather than sequence-driven structure, and comparing the two arms separates the two. v1's files are carried forward unchanged; this is a pure addition.New files: rwinit_core.tar.gz (analysis outputs: descriptors, embedding and downstream comparisons between the two arms), rwinit_images.tar.gz (528x3x9 KDE-rendered images), and rwinit_gsd_shard_01..rwinit_gsd_shard_16 (1584 GSD trajectories, ~57 MB per shard), plus manifest_rwinit.json and sequence_to_shard_rwinit.csv.Three things to note when using both arms together. (1) The two arms use different sequence-to-shard assignments: both partition the same 528 sequences into 16 shards of 33, but 480 sequences fall in a different shard number between the arms. Use sequence_to_shard_rwinit.csv for the rwinit arm; reusing sequence_to_shard.csv will fetch a shard that does not contain the sequence. (2) The archives extract to distinct roots (kg10-gsd-bundle-rwinit/, exhaustive10-rwinit/) so both arms can be unpacked side by side without overwriting each other, since the per-file names are identical between arms. (3) rwinit_images.tar.gz ships as WebP-lossless, matching v1's image format and verified bit-identical to the rendered PNGs; the analysis scripts archived inside rwinit_core.tar.gz refer to .png because they were run against those PNG originals before conversion.