--- configs: - config_name: chains data_files: - split: train path: chains/train-* - split: valid path: chains/valid-* - split: test path: chains/test-* - config_name: complexes data_files: - split: train path: complexes/train-* - split: valid path: complexes/valid-* - split: test path: complexes/test-* --- # PDB-Chain-Complex-Benchmark-Rigor Rigor rebuild of `Synthyra/PDB-Chain-Complex-Benchmark` with split assignments recomputed from the published chain and complex parquet artifacts. ## Split Policy Splits are assigned by connected components over exact sequence, sequence hash, 30% sequence cluster, structure cluster, source split component, same-PDB asymmetric-unit membership, chain assembly membership, and biological assembly co-membership from the complex rows. The target is 90/5/5 by component count. The rebuild hard-fails before upload if any chain or complex member leaks across train, validation, and test by chain UID, sequence, sequence hash, 30% sequence cluster, structure cluster, or recomputed split component. Chains shorter than `min_sequence_length` or containing `X` residues when `exclude_sequences_with_x=true` are removed before graph construction. Biological assembly rows containing any removed member chain are dropped rather than published as partial complexes. Complex list-valued payload columns are regenerated from the chain table during the rebuild, then asserted directly from the emitted payloads, including `asymmetric_unit_chain_uids`. After writing parquet, the builder reloads the emitted chain and complex shards and reruns split-routing, payload-consistency, and disjointness checks from the serialized files. This artifact-only rebuild does not rerun Foldseek or MMseqs2. It enforces and verifies the published cluster labels plus optional external sequence verification edges when supplied. ## Cluster-Balanced PyTorch Sampling The training split can be loaded with `streaming=False` and sampled as one random row per sequence-cluster group per epoch. Chain mode groups by `sequence_cluster_30`. Complex mode defaults to one sample per member sequence cluster from `sequence_clusters_30`; set `complex_grouping="cluster_set"` to sample one row per unique complex cluster signature instead. ```python import collections import random from typing import Any, Mapping, Optional from datasets import load_dataset from torch.utils.data import DataLoader, Dataset DEFAULT_REPO_ID = "Synthyra/PDB-Chain-Complex-Benchmark-Rigor" class PDBClusteredDataset(Dataset): """One randomly selected row per sequence-cluster group per epoch.""" def __init__( self, repo_id: str = DEFAULT_REPO_ID, mode: str = "chains", split: str = "train", seed: int = 0, epoch: int = 0, cluster_key: Optional[str] = None, complex_grouping: str = "member_cluster", load_kwargs: Optional[Mapping[str, Any]] = None, ) -> None: mode_to_config = { "single": "chains", "chains": "chains", "complex": "complexes", "complexes": "complexes", } self.config = mode_to_config[mode] self.seed = int(seed) self.cluster_key = cluster_key or ( "sequence_cluster_30" if self.config == "chains" else "sequence_clusters_30" ) self.complex_grouping = complex_grouping self.dataset = load_dataset( repo_id, self.config, split=split, streaming=False, **dict(load_kwargs or {}), ) self.groups = self._build_groups() self.group_keys = sorted(self.groups) self.set_epoch(epoch) def _group_keys_for_value(self, value: Any) -> tuple[tuple[str, ...], ...]: if self.config == "chains": return ((str(value),),) clusters = tuple(sorted({str(cluster) for cluster in value if str(cluster)})) if not clusters: raise ValueError("complex row has no sequence cluster labels") if self.complex_grouping == "cluster_set": return (clusters,) if self.complex_grouping == "member_cluster": return tuple((cluster,) for cluster in clusters) raise ValueError("complex_grouping must be 'member_cluster' or 'cluster_set'") def _build_groups(self) -> dict[tuple[str, ...], list[int]]: groups = collections.defaultdict(list) for row_idx, value in enumerate(self.dataset[self.cluster_key]): for key in self._group_keys_for_value(value): groups[key].append(row_idx) if not groups: raise ValueError("dataset split contains no rows") return dict(groups) def set_epoch(self, epoch: int) -> None: self.epoch = int(epoch) rng = random.Random(self.seed + self.epoch) self.indices = [rng.choice(self.groups[key]) for key in self.group_keys] def __len__(self) -> int: return len(self.indices) def __getitem__(self, index: int) -> Mapping[str, Any]: return self.dataset[self.indices[index]] def identity_collate(batch: list[Mapping[str, Any]]) -> list[Mapping[str, Any]]: return batch chains = PDBClusteredDataset(mode="single", split="train", seed=42) complexes = PDBClusteredDataset(mode="complex", split="train", seed=42) chains_loader = DataLoader(chains, batch_size=8, shuffle=False, collate_fn=identity_collate) complex_loader = DataLoader(complexes, batch_size=2, shuffle=False, collate_fn=identity_collate) for epoch in range(3): chains.set_epoch(epoch) complexes.set_epoch(epoch) first_chain_batch = next(iter(chains_loader)) first_complex_batch = next(iter(complex_loader)) ``` ## Build Metadata ```json { "batch_size": 16384, "chain_scan": { "assembly_membership_edges_seen": 1105293, "asymmetric_unit_membership_edges_seen": 1037226, "chains_excluded_contains_x": 24118, "chains_excluded_too_short": 13084, "chains_excluded_total": 32222, "chains_retained": 1037226, "chains_scanned": 1069448 }, "chain_write": { "file_counts": { "test": 1, "train": 21, "valid": 1 }, "row_counts": { "test": 302, "train": 1036622, "valid": 302 } }, "complex_edge_scan": { "complex_member_edges_seen": 642454, "complexes_dropped_by_filtered_member": 18838, "complexes_scanned_for_edges": 344745, "source_complex_rows_seen": 363583 }, "complex_write": { "complexes_dropped_by_filtered_member": 18838, "file_counts": { "test": 1, "train": 7, "valid": 1 }, "member_chain_rows": 987199, "member_fields": { "asymmetric_unit_chain_uids": { "crossing_values": 0, "max_pairwise_overlap": 0, "unique_values": 923757 }, "chain_uid": { "crossing_values": 0, "max_pairwise_overlap": 0, "unique_values": 921450 }, "sequence_clusters_30": { "crossing_values": 0, "max_pairwise_overlap": 0, "unique_values": 41718 }, "sequence_hashes": { "crossing_values": 0, "max_pairwise_overlap": 0, "unique_values": 166930 }, "sequences": { "crossing_values": 0, "max_pairwise_overlap": 0, "unique_values": 166930 }, "split_components": { "crossing_values": 0, "max_pairwise_overlap": 0, "unique_values": 6019 }, "structure_clusters": { "crossing_values": 0, "max_pairwise_overlap": 0, "unique_values": 44485 } }, "row_counts": { "test": 301, "train": 344140, "valid": 304 } }, "component_stats": { "split_component_count": 6033, "test_component_count": 302, "train_component_count": 5429, "valid_component_count": 302 }, "cpu_count": 32, "dropped_complex_membership_keys": 18838, "elapsed_seconds": 77.882, "exclude_sequences_with_x": true, "filtered_chain_rows": 32222, "filtered_complex_rows": 18838, "foldseek_path": "", "include_source_split_component_edges": true, "max_rows_per_shard": 50000, "min_sequence_length": 10, "mmseqs_path": "", "nvidia-smi_path": "C:\\Windows\\system32\\nvidia-smi.EXE", "nvidia-smi_version": "Mon Jul 6 14:22:11 2026 \n+-----------------------------------------------------------------------------------------+\n| NVIDIA-SMI 580.88 Driver Version: 580.88 CUDA Version: 13.0 |\n+-----------------------------------------+------------------------+----------------------+\n| GPU Name Driver-Model | Bus-Id Disp.A | Volatile Uncorr. ECC |\n| Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. |\n| | | MIG M. |\n|=========================================+========================+======================|\n| 0 NVIDIA GeForce RTX 4070 ... WDDM | 00000000:01:00.0 On | N/A |\n| N/A 43C P5 8W / 123W | 866MiB / 8188MiB | 26% Default |\n| | | N/A |\n+-----------------------------------------+------------------------+----------------------+\n\n+-----------------------------------------------------------------------------------------+\n| Processes: |\n| GPU GI CI PID Type Process name GPU Memory |\n| ID ID Usage |\n|=========================================================================================|\n| 0 N/A N/A 6516 C+G ...p\\app-3.6.2\\GitHubDesktop.exe N/A |\n| 0 N/A N/A 8844 C+G ...indows\\System32\\ShellHost.exe N/A |\n| 0 N/A N/A 10456 C+G ..._cw5n1h2txyewy\\SearchHost.exe N/A |\n| 0 N/A N/A 13212 C+G ...5n1h2txyewy\\TextInputHost.exe N/A |\n| 0 N/A N/A 21252 C+G ...__2p2nqsd0c76g0\\app\\Codex.exe N/A |\n| 0 N/A N/A 36048 C+G ...Chrome\\Application\\chrome.exe N/A |\n| 0 N/A N/A 41088 C+G ...ntrolPanel\\SystemSettings.exe N/A |\n| 0 N/A N/A 63696 C+G ...8bbwe\\PhoneExperienceHost.exe N/A |\n| 0 N/A N/A 72372 C+G ...y\\StartMenuExperienceHost.exe N/A |\n| 0 N/A N/A 84244 C+G ...xyewy\\ShellExperienceHost.exe N/A |\n| 0 N/A N/A 104536 C+G C:\\Windows\\explorer.exe N/A |\n| 0 N/A N/A 152648 C+G ....0.4022.80\\msedgewebview2.exe N/A |\n| 0 N/A N/A 152984 C+G ...e Experience\\NVIDIA Share.exe N/A |\n| 0 N/A N/A 155976 C+G ...em32\\ApplicationFrameHost.exe N/A |\n| 0 N/A N/A 196292 C+G ...Chrome\\Application\\chrome.exe N/A |\n| 0 N/A N/A 206536 C+G ...2txyewy\\CrossDeviceResume.exe N/A |\n| 0 N/A N/A 221536 C+G ...Files\\Notepad++\\notepad++.exe N/A |\n+-----------------------------------------------------------------------------------------+", "platform": "Windows-11-10.0.26200-SP0", "python": "3.12.13", "quality_filter": { "drop_complexes_with_filtered_members": true, "exclude_sequences_with_x": true, "min_sequence_length": 10 }, "retained_chain_rows": 1037226, "retained_complex_rows": 344745, "seed": 42, "sequence_verification_max_unknown_fraction": 0.0, "source_chain_rows": 1069448, "source_complex_rows": 363583, "source_repo_id": "Synthyra/PDB-Chain-Complex-Benchmark", "source_snapshot_dir": "C:\\tmp\\hf_pdb_chain_complex_cache\\datasets--Synthyra--PDB-Chain-Complex-Benchmark\\snapshots\\3eb55ee7289dfedfbddf0cde48b68c5395783ccf", "target_repo_id": "Synthyra/PDB-Chain-Complex-Benchmark-Rigor", "test_frac": 0.05, "valid_frac": 0.05 } ``` ## Leakage Assertions ```json { "chains": { "fields": { "chain_uid": { "crossing_values": 0, "max_pairwise_overlap": 0, "unique_values": 1037226 }, "seqs": { "crossing_values": 0, "max_pairwise_overlap": 0, "unique_values": 175111 }, "sequence_cluster_30": { "crossing_values": 0, "max_pairwise_overlap": 0, "unique_values": 43208 }, "sequence_hash": { "crossing_values": 0, "max_pairwise_overlap": 0, "unique_values": 175111 }, "split_component": { "crossing_values": 0, "max_pairwise_overlap": 0, "unique_values": 6033 }, "structure_cluster": { "crossing_values": 0, "max_pairwise_overlap": 0, "unique_values": 47196 } }, "post_write": { "fields": { "chain_uid": { "crossing_values": 0, "max_pairwise_overlap": 0, "unique_values": 1037226 }, "seqs": { "crossing_values": 0, "max_pairwise_overlap": 0, "unique_values": 175111 }, "sequence_cluster_30": { "crossing_values": 0, "max_pairwise_overlap": 0, "unique_values": 43208 }, "sequence_hash": { "crossing_values": 0, "max_pairwise_overlap": 0, "unique_values": 175111 }, "split_component": { "crossing_values": 0, "max_pairwise_overlap": 0, "unique_values": 6033 }, "structure_cluster": { "crossing_values": 0, "max_pairwise_overlap": 0, "unique_values": 47196 } }, "rows": 1037226, "split_counts": { "test": 302, "train": 1036622, "valid": 302 } }, "rows": 1037226, "sequence_verification_edges": { "checked_edges": 0, "cross_split_edges": 0, "provided": false, "skipped_self_edges": 0, "skipped_unknown_edges": 0, "total_edges": 0 }, "split_counts": { "test": 302, "train": 1036622, "valid": 302 } }, "complexes": { "member_chain_rows": 987199, "member_fields": { "asymmetric_unit_chain_uids": { "crossing_values": 0, "max_pairwise_overlap": 0, "unique_values": 923757 }, "chain_uid": { "crossing_values": 0, "max_pairwise_overlap": 0, "unique_values": 921450 }, "sequence_clusters_30": { "crossing_values": 0, "max_pairwise_overlap": 0, "unique_values": 41718 }, "sequence_hashes": { "crossing_values": 0, "max_pairwise_overlap": 0, "unique_values": 166930 }, "sequences": { "crossing_values": 0, "max_pairwise_overlap": 0, "unique_values": 166930 }, "split_components": { "crossing_values": 0, "max_pairwise_overlap": 0, "unique_values": 6019 }, "structure_clusters": { "crossing_values": 0, "max_pairwise_overlap": 0, "unique_values": 44485 } }, "member_split_consistency": "passed", "payload_columns_regenerated_from_chains": true, "post_write": { "asymmetric_unit_chain_rows": 1541919, "member_chain_rows": 987199, "member_fields": { "asymmetric_unit_chain_uids": { "crossing_values": 0, "max_pairwise_overlap": 0, "unique_values": 923757 }, "chain_uid": { "crossing_values": 0, "max_pairwise_overlap": 0, "unique_values": 921450 }, "sequence_clusters_30": { "crossing_values": 0, "max_pairwise_overlap": 0, "unique_values": 41718 }, "sequence_hashes": { "crossing_values": 0, "max_pairwise_overlap": 0, "unique_values": 166930 }, "sequences": { "crossing_values": 0, "max_pairwise_overlap": 0, "unique_values": 166930 }, "split_components": { "crossing_values": 0, "max_pairwise_overlap": 0, "unique_values": 6019 }, "structure_clusters": { "crossing_values": 0, "max_pairwise_overlap": 0, "unique_values": 44485 } }, "payload_columns_match_chain_rows": true, "rows": 344745, "shard_split_routing": "passed", "split_counts": { "test": 301, "train": 344140, "valid": 304 } }, "rows": 344745, "split_counts": { "test": 301, "train": 344140, "valid": 304 } } } ```