RosettaCommons/MAHOMES_II

MAHOMES II

MAHOMES-II (Metal Activity Heuristic of Metalloprotein and Enzymatic Sites-II) is a structure-based dataset for classifying protein-bound metal sites as enzymatic or non-enzymatic. ('Enzyme' column)

Dataset Splits

The train/test split follows a temporal split based on PDB deposition date. Structures that were deposited before 2018 were included in the training dataset, and structures deposited in 2018 or later went into the holdout test set. The train dataset contains 3522 entries (974 enzyme sites and 2548 nonenzyme sites). The test set consists of 5285 entries (1795 enzyme sites and 3490 nonenzyme sites). The test set was augmented by running multiple Rosetta relaxation replicates per structure, thus resulting in the higher number of entries, while the train set only took the best relaxation structure. Intra-set and inter-set redundancy were removed using sequence homology (PHMMER) and local structural similarity (Jaccard index of coordinating residues).

The paper states that: The train dataset contains 3424 entries (957 enzyme sites and 2467 nonenzyme sites). The test set consists of 5172 entries (1895 enzyme sites and 3277 nonenzyme sites), which were representative of 517 entries (189 enzyme sites and 328 nonenzyme sites).

Quickstart Usage

Install HuggingFace Datasets package

Each subset can be loaded into python using the Huggingface datasets library. First, from the command line install the datasets library

$ pip install datasets

then, from within python load the datasets library

>>> import datasets

Load dataset

Load the 'MAHOMES_II' datasets.

>>> MAHOMES_II = datasets.load_dataset("RosettaCommons/MAHOMES_II", name = "default")                     
Downloading readme: 2.23kB [00:00, 11.1MB/s]
Downloading data: 100%|███████████████████████████████████████████████████████████████████████████████████████████| 7.15M/7.15M [00:00<00:00, 20.7MB/s]
Downloading data: 100%|███████████████████████████████████████████████████████████████████████████████████████████| 11.0M/11.0M [00:00<00:00, 30.7MB/s]
Generating train split: 100%|████████████████████████████████████████████████████████████████████████████| 4866/4866 [00:00<00:00, 41639.85 examples/s]
Generating test split: 100%|█████████████████████████████████████████████████████████████████████████████| 7844/7844 [00:00<00:00, 44324.19 examples/s]

and the dataset is loaded as a datasets.arrow_dataset.Dataset

>>> MAHOMES_II
DatasetDict({
       train: Dataset({
       features: ['SITE_ID', 'struc_id', 'MetalCodes', 'MetalAtoms', 'SiteAtoms', 'bad_site', 'note', 'metal1_atom_name', 'metal1_serial', 'metal1_resName', 'metal1_seqID', 'Elec_ins_mean_mu2', 'Elec_ins_mean_mu3', 'Elec_ins_mean_mu4', 'Elec_ins_mean_GBR6', 'Elec_ins_mean_pKa_shift', 'Elec_ins_mean_dpKa_titr', 'Elec_ins_mean_dpKa_desolv', 'Elec_ins_mean_dpKa_bg', 'Elec_ins_mean_DestabRank', 'Elec_ins_mean_StabRank', 'Elec_ins_mean_ResSigDev', 'Elec_ins_mean_SolvEnergy', 'Elec_ins_mean_SolvExp', 'Elec_ins_std_mu2', 'Elec_ins_std_mu3', 'Elec_ins_std_mu4', 'Elec_ins_std_GBR6', 'Elec_ins_std_pKa_shift', 'Elec_ins_std_dpKa_titr', 'Elec_ins_std_dpKa_desolv', 'Elec_ins_std_dpKa_bg', 'Elec_ins_std_DestabRank', 'Elec_ins_std_StabRank', 'Elec_ins_std_ResSigDev', 'Elec_ins_std_SolvEnergy', 'Elec_ins_std_SolvExp', 'Elec_ins_min_mu2', 'Elec_ins_min_mu3', 'Elec_ins_min_mu4', 'Elec_ins_min_GBR6', 'Elec_ins_min_pKa_shift', 'Elec_ins_min_dpKa_titr', 'Elec_ins_min_dpKa_desolv', 'Elec_ins_min_dpKa_bg', 'Elec_ins_min_DestabRank', 'Elec_ins_min_StabRank', 'Elec_ins_min_ResSigDev', 'Elec_ins_min_SolvEnergy', 'Elec_ins_min_SolvExp', 'Elec_ins_max_mu2', 'Elec_ins_max_mu3', 'Elec_ins_max_mu4', 'Elec_ins_max_GBR6', 'Elec_ins_max_pKa_shift', 'Elec_ins_max_dpKa_titr', 'Elec_ins_max_dpKa_desolv', 'Elec_ins_max_dpKa_bg', 'Elec_ins_max_DestabRank', 'Elec_ins_max_StabRank', 'Elec_ins_max_ResSigDev', 'Elec_ins_max_SolvEnergy', 'Elec_ins_max_SolvExp', 'Elec_ins_range_mu2', 'Elec_ins_range_mu3', 'Elec_ins_range_mu4', 'Elec_ins_range_GBR6', 'Elec_ins_range_pKa_shift', 'Elec_ins_range_dpKa_titr', 'Elec_ins_range_dpKa_desolv', 'Elec_ins_range_dpKa_bg', 'Elec_ins_range_DestabRank', 'Elec_ins_range_StabRank', 'Elec_ins_range_ResSigDev', 'Elec_ins_range_SolvEnergy', 'Elec_ins_range_SolvExp', 'Elec_ins_Z_mu2_count', 'Elec_ins_Z_mu3_count', 'Elec_ins_Z_mu4_count', 'Elec_outs_mean_mu2', 'Elec_outs_mean_mu3', 'Elec_outs_mean_mu4', 'Elec_outs_mean_GBR6', 'Elec_outs_mean_pKa_shift', 'Elec_outs_mean_dpKa_titr', 'Elec_outs_mean_dpKa_desolv', 'Elec_outs_mean_dpKa_bg', 'Elec_outs_mean_DestabRank', 'Elec_outs_mean_StabRank', 'Elec_outs_mean_ResSigDev', 'Elec_outs_mean_SolvEnergy', 'Elec_outs_mean_SolvExp', 'Elec_outs_std_mu2', 'Elec_outs_std_mu3', 'Elec_outs_std_mu4', 'Elec_outs_std_GBR6', 'Elec_outs_std_pKa_shift', 'Elec_outs_std_dpKa_titr', 'Elec_outs_std_dpKa_desolv', 'Elec_outs_std_dpKa_bg', 'Elec_outs_std_DestabRank', 'Elec_outs_std_StabRank', 'Elec_outs_std_ResSigDev', 'Elec_outs_std_SolvEnergy', 'Elec_outs_std_SolvExp', 'Elec_outs_min_mu2', 'Elec_outs_min_mu3', 'Elec_outs_min_mu4', 'Elec_outs_min_GBR6', 'Elec_outs_min_pKa_shift', 'Elec_outs_min_dpKa_titr', 'Elec_outs_min_dpKa_desolv', 'Elec_outs_min_dpKa_bg', 'Elec_outs_min_DestabRank', 'Elec_outs_min_StabRank', 'Elec_outs_min_ResSigDev', 'Elec_outs_min_SolvEnergy', 'Elec_outs_min_SolvExp', 'Elec_outs_max_mu2', 'Elec_outs_max_mu3', 'Elec_outs_max_mu4', 'Elec_outs_max_GBR6', 'Elec_outs_max_pKa_shift', 'Elec_outs_max_dpKa_titr', 'Elec_outs_max_dpKa_desolv', 'Elec_outs_max_dpKa_bg', 'Elec_outs_max_DestabRank', 'Elec_outs_max_StabRank', 'Elec_outs_max_ResSigDev', 'Elec_outs_max_SolvEnergy', 'Elec_outs_max_SolvExp', 'Elec_outs_range_mu2', 'Elec_outs_range_mu3', 'Elec_outs_range_mu4', 'Elec_outs_range_GBR6', 'Elec_outs_range_pKa_shift', 'Elec_outs_range_dpKa_titr', 'Elec_outs_range_dpKa_desolv', 'Elec_outs_range_dpKa_bg', 'Elec_outs_range_DestabRank', 'Elec_outs_range_StabRank', 'Elec_outs_range_ResSigDev', 'Elec_outs_range_SolvEnergy', 'Elec_outs_range_SolvExp', 'Elec_outs_Z_mu2_count', 'Elec_outs_Z_mu3_count', 'Elec_outs_Z_mu4_count', 'Elec_xenv_mu2', 'Elec_xenv_mu3', 'Elec_xenv_mu4', 'Elec_xenv_GBR6', 'Elec_xenv_pKa_shift', 'Elec_xenv_dpKa_titr', 'Elec_xenv_dpKa_desolv', 'Elec_xenv_dpKa_bg', 'Elec_xenv_DestabRank', 'Elec_xenv_StabRank', 'Elec_xenv_ResSigDev', 'Elec_xenv_SolvEnergy', 'Elec_xenv_SolvExp', 'Elec_ins_numResis', 'Elec_outs_numResis', 'Elec_xenvr_numResis', 'geom_lin', 'geom_trv', 'geom_tri', 'geom_tev', 'geom_spv', 'geom_tet', 'geom_spl', 'geom_bva', 'geom_bvp', 'geom_pyv', 'geom_spy', 'geom_tbp', 'geom_tpv', 'geom_oct', 'geom_tpr', 'geom_pva', 'geom_pvp', 'geom_cof', 'geom_con', 'geom_ctf', 'geom_ctn', 'geom_pbp', 'geom_coc', 'geom_ctp', 'geom_hva', 'geom_hvp', 'geom_cuv', 'geom_sav', 'geom_hbp', 'geom_cub', 'geom_sqa', 'geom_boc', 'geom_bts', 'geom_btt', 'geom_ttp', 'geom_csa', 'geom_irr', 'geom_Reg', 'geom_Distort', 'geom_LigN', 'geom_LigO', 'geom_LigS', 'geom_LigOther', 'geom_AtomRMSD', 'geom_cn2', 'geom_cn3', 'geom_cn4', 'geom_cn5', 'geom_cn6', 'geom_cn7', 'geom_cn8', 'geom_cn9', 'geom_Filled', 'geom_PartFilled', 'geom_AvgN', 'geom_AvgO', 'geom_AvgS', 'geom_AvgOther', 'rbf_fa_atr', 'rbf_fa_rep', 'rbf_fa_sol', 'rbf_fa_intra_atr_xover4', 'rbf_fa_intra_rep_xover4', 'rbf_fa_intra_sol_xover4', 'rbf_lk_ball', 'rbf_lk_ball_iso', 'rbf_lk_ball_bridge', 'rbf_lk_ball_bridge_uncpl', 'rbf_fa_elec', 'rbf_fa_intra_elec', 'rbf_pro_close', 'rbf_hbond_sr_bb', 'rbf_hbond_lr_bb', 'rbf_hbond_bb_sc', 'rbf_hbond_sc', 'rbf_dslf_fa13', 'rbf_omega', 'rbf_fa_dun', 'rbf_fa_dun_dev', 'rbf_fa_dun_rot', 'rbf_fa_dun_semi', 'rbf_p_aa_pp', 'rbf_yhh_planarity', 'rbf_hxl_tors', 'rbf_ref', 'rbf_rama_prepro', 'rbf_total', 'rbf_numScoredResis', 'pkt_shape_Depth', 'pkt_shape_Vol', 'pkt_shape_LongPath', 'pkt_shape_farPtLow', 'pkt_shape_PocketAreaLow', 'pkt_shape_OffsetLow', 'pkt_shape_LongAxLow', 'pkt_shape_ShortAxLow', 'pkt_shape_farPtMid', 'pkt_shape_PocketAreaMid', 'pkt_shape_OffsetMid', 'pkt_shape_LongAxMid', 'pkt_shape_ShortAxMid', 'pkt_shape_farPtHigh', 'pkt_shape_PocketAreaHigh', 'pkt_shape_OffsetHigh', 'pkt_shape_LongAxHigh', 'pkt_shape_ShortAxHigh', 'pkt_shape_SITEDistCenter', 'pkt_shape_SITEDistNormCenter', 'pkt_lining_in_pocket', 'pkt_lining_num_pocket_bb', 'pkt_lining_num_pocket_sc', 'pkt_lining_avg_eisen_hp', 'pkt_lining_min_eisen', 'pkt_lining_max_eisen', 'pkt_lining_skew_eisen', 'pkt_lining_std_dev_eisen', 'pkt_lining_avg_kyte_hp', 'pkt_lining_min_kyte', 'pkt_lining_max_kyte', 'pkt_lining_skew_kyte', 'pkt_lining_std_dev_kyte', 'pkt_lining_occ_vol', 'pkt_lining_NoSC_vol', 'pkt_lining_SC_vol_perc', 'pkt_info_ClusterNum', 'pkt_info_Ngrid', 'pkt_info_Volume', 'pkt_info_Rinac(A) av', 'pkt_info_Rinac(A) mi', 'pkt_info_invRvolume(AA)', 'pkt_info_SITE_pocket_distance_min', 'pkt_info_num_adjacent_pockets', 'pkt_info_pocket_height', 'pkt_info_pocket_depth', 'pkt_info_metal_height', 'pkt_info_metal_depth', 'metal2_atom_name', 'metal2_serial', 'metal2_resName', 'metal2_seqID', 'metal3_atom_name', 'metal3_serial', 'metal3_resName', 'metal3_seqID', 'metal4_atom_name', 'metal4_serial', 'metal4_resName', 'metal4_seqID', 'Enzyme'],
       num_rows: 4866
    })
    test: Dataset({
       features: ['SITE_ID', 'struc_id', 'MetalCodes', 'MetalAtoms', 'SiteAtoms', 'bad_site', 'note', 'metal1_atom_name', 'metal1_serial', 'metal1_resName', 'metal1_seqID', 'Elec_ins_mean_mu2', 'Elec_ins_mean_mu3', 'Elec_ins_mean_mu4', 'Elec_ins_mean_GBR6', 'Elec_ins_mean_pKa_shift', 'Elec_ins_mean_dpKa_titr', 'Elec_ins_mean_dpKa_desolv', 'Elec_ins_mean_dpKa_bg', 'Elec_ins_mean_DestabRank', 'Elec_ins_mean_StabRank', 'Elec_ins_mean_ResSigDev', 'Elec_ins_mean_SolvEnergy', 'Elec_ins_mean_SolvExp', 'Elec_ins_std_mu2', 'Elec_ins_std_mu3', 'Elec_ins_std_mu4', 'Elec_ins_std_GBR6', 'Elec_ins_std_pKa_shift', 'Elec_ins_std_dpKa_titr', 'Elec_ins_std_dpKa_desolv', 'Elec_ins_std_dpKa_bg', 'Elec_ins_std_DestabRank', 'Elec_ins_std_StabRank', 'Elec_ins_std_ResSigDev', 'Elec_ins_std_SolvEnergy', 'Elec_ins_std_SolvExp', 'Elec_ins_min_mu2', 'Elec_ins_min_mu3', 'Elec_ins_min_mu4', 'Elec_ins_min_GBR6', 'Elec_ins_min_pKa_shift', 'Elec_ins_min_dpKa_titr', 'Elec_ins_min_dpKa_desolv', 'Elec_ins_min_dpKa_bg', 'Elec_ins_min_DestabRank', 'Elec_ins_min_StabRank', 'Elec_ins_min_ResSigDev', 'Elec_ins_min_SolvEnergy', 'Elec_ins_min_SolvExp', 'Elec_ins_max_mu2', 'Elec_ins_max_mu3', 'Elec_ins_max_mu4', 'Elec_ins_max_GBR6', 'Elec_ins_max_pKa_shift', 'Elec_ins_max_dpKa_titr', 'Elec_ins_max_dpKa_desolv', 'Elec_ins_max_dpKa_bg', 'Elec_ins_max_DestabRank', 'Elec_ins_max_StabRank', 'Elec_ins_max_ResSigDev', 'Elec_ins_max_SolvEnergy', 'Elec_ins_max_SolvExp', 'Elec_ins_range_mu2', 'Elec_ins_range_mu3', 'Elec_ins_range_mu4', 'Elec_ins_range_GBR6', 'Elec_ins_range_pKa_shift', 'Elec_ins_range_dpKa_titr', 'Elec_ins_range_dpKa_desolv', 'Elec_ins_range_dpKa_bg', 'Elec_ins_range_DestabRank', 'Elec_ins_range_StabRank', 'Elec_ins_range_ResSigDev', 'Elec_ins_range_SolvEnergy', 'Elec_ins_range_SolvExp', 'Elec_ins_Z_mu2_count', 'Elec_ins_Z_mu3_count', 'Elec_ins_Z_mu4_count', 'Elec_outs_mean_mu2', 'Elec_outs_mean_mu3', 'Elec_outs_mean_mu4', 'Elec_outs_mean_GBR6', 'Elec_outs_mean_pKa_shift', 'Elec_outs_mean_dpKa_titr', 'Elec_outs_mean_dpKa_desolv', 'Elec_outs_mean_dpKa_bg', 'Elec_outs_mean_DestabRank', 'Elec_outs_mean_StabRank', 'Elec_outs_mean_ResSigDev', 'Elec_outs_mean_SolvEnergy', 'Elec_outs_mean_SolvExp', 'Elec_outs_std_mu2', 'Elec_outs_std_mu3', 'Elec_outs_std_mu4', 'Elec_outs_std_GBR6', 'Elec_outs_std_pKa_shift', 'Elec_outs_std_dpKa_titr', 'Elec_outs_std_dpKa_desolv', 'Elec_outs_std_dpKa_bg', 'Elec_outs_std_DestabRank', 'Elec_outs_std_StabRank', 'Elec_outs_std_ResSigDev', 'Elec_outs_std_SolvEnergy', 'Elec_outs_std_SolvExp', 'Elec_outs_min_mu2', 'Elec_outs_min_mu3', 'Elec_outs_min_mu4', 'Elec_outs_min_GBR6', 'Elec_outs_min_pKa_shift', 'Elec_outs_min_dpKa_titr', 'Elec_outs_min_dpKa_desolv', 'Elec_outs_min_dpKa_bg', 'Elec_outs_min_DestabRank', 'Elec_outs_min_StabRank', 'Elec_outs_min_ResSigDev', 'Elec_outs_min_SolvEnergy', 'Elec_outs_min_SolvExp', 'Elec_outs_max_mu2', 'Elec_outs_max_mu3', 'Elec_outs_max_mu4', 'Elec_outs_max_GBR6', 'Elec_outs_max_pKa_shift', 'Elec_outs_max_dpKa_titr', 'Elec_outs_max_dpKa_desolv', 'Elec_outs_max_dpKa_bg', 'Elec_outs_max_DestabRank', 'Elec_outs_max_StabRank', 'Elec_outs_max_ResSigDev', 'Elec_outs_max_SolvEnergy', 'Elec_outs_max_SolvExp', 'Elec_outs_range_mu2', 'Elec_outs_range_mu3', 'Elec_outs_range_mu4', 'Elec_outs_range_GBR6', 'Elec_outs_range_pKa_shift', 'Elec_outs_range_dpKa_titr', 'Elec_outs_range_dpKa_desolv', 'Elec_outs_range_dpKa_bg', 'Elec_outs_range_DestabRank', 'Elec_outs_range_StabRank', 'Elec_outs_range_ResSigDev', 'Elec_outs_range_SolvEnergy', 'Elec_outs_range_SolvExp', 'Elec_outs_Z_mu2_count', 'Elec_outs_Z_mu3_count', 'Elec_outs_Z_mu4_count', 'Elec_xenv_mu2', 'Elec_xenv_mu3', 'Elec_xenv_mu4', 'Elec_xenv_GBR6', 'Elec_xenv_pKa_shift', 'Elec_xenv_dpKa_titr', 'Elec_xenv_dpKa_desolv', 'Elec_xenv_dpKa_bg', 'Elec_xenv_DestabRank', 'Elec_xenv_StabRank', 'Elec_xenv_ResSigDev', 'Elec_xenv_SolvEnergy', 'Elec_xenv_SolvExp', 'Elec_ins_numResis', 'Elec_outs_numResis', 'Elec_xenvr_numResis', 'geom_lin', 'geom_trv', 'geom_tri', 'geom_tev', 'geom_spv', 'geom_tet', 'geom_spl', 'geom_bva', 'geom_bvp', 'geom_pyv', 'geom_spy', 'geom_tbp', 'geom_tpv', 'geom_oct', 'geom_tpr', 'geom_pva', 'geom_pvp', 'geom_cof', 'geom_con', 'geom_ctf', 'geom_ctn', 'geom_pbp', 'geom_coc', 'geom_ctp', 'geom_hva', 'geom_hvp', 'geom_cuv', 'geom_sav', 'geom_hbp', 'geom_cub', 'geom_sqa', 'geom_boc', 'geom_bts', 'geom_btt', 'geom_ttp', 'geom_csa', 'geom_irr', 'geom_Reg', 'geom_Distort', 'geom_LigN', 'geom_LigO', 'geom_LigS', 'geom_LigOther', 'geom_AtomRMSD', 'geom_cn2', 'geom_cn3', 'geom_cn4', 'geom_cn5', 'geom_cn6', 'geom_cn7', 'geom_cn8', 'geom_cn9', 'geom_Filled', 'geom_PartFilled', 'geom_AvgN', 'geom_AvgO', 'geom_AvgS', 'geom_AvgOther', 'rbf_fa_atr', 'rbf_fa_rep', 'rbf_fa_sol', 'rbf_fa_intra_atr_xover4', 'rbf_fa_intra_rep_xover4', 'rbf_fa_intra_sol_xover4', 'rbf_lk_ball', 'rbf_lk_ball_iso', 'rbf_lk_ball_bridge', 'rbf_lk_ball_bridge_uncpl', 'rbf_fa_elec', 'rbf_fa_intra_elec', 'rbf_pro_close', 'rbf_hbond_sr_bb', 'rbf_hbond_lr_bb', 'rbf_hbond_bb_sc', 'rbf_hbond_sc', 'rbf_dslf_fa13', 'rbf_omega', 'rbf_fa_dun', 'rbf_fa_dun_dev', 'rbf_fa_dun_rot', 'rbf_fa_dun_semi', 'rbf_p_aa_pp', 'rbf_yhh_planarity', 'rbf_hxl_tors', 'rbf_ref', 'rbf_rama_prepro', 'rbf_total', 'rbf_numScoredResis', 'pkt_shape_Depth', 'pkt_shape_Vol', 'pkt_shape_LongPath', 'pkt_shape_farPtLow', 'pkt_shape_PocketAreaLow', 'pkt_shape_OffsetLow', 'pkt_shape_LongAxLow', 'pkt_shape_ShortAxLow', 'pkt_shape_farPtMid', 'pkt_shape_PocketAreaMid', 'pkt_shape_OffsetMid', 'pkt_shape_LongAxMid', 'pkt_shape_ShortAxMid', 'pkt_shape_farPtHigh', 'pkt_shape_PocketAreaHigh', 'pkt_shape_OffsetHigh', 'pkt_shape_LongAxHigh', 'pkt_shape_ShortAxHigh', 'pkt_shape_SITEDistCenter', 'pkt_shape_SITEDistNormCenter', 'pkt_lining_in_pocket', 'pkt_lining_num_pocket_bb', 'pkt_lining_num_pocket_sc', 'pkt_lining_avg_eisen_hp', 'pkt_lining_min_eisen', 'pkt_lining_max_eisen', 'pkt_lining_skew_eisen', 'pkt_lining_std_dev_eisen', 'pkt_lining_avg_kyte_hp', 'pkt_lining_min_kyte', 'pkt_lining_max_kyte', 'pkt_lining_skew_kyte', 'pkt_lining_std_dev_kyte', 'pkt_lining_occ_vol', 'pkt_lining_NoSC_vol', 'pkt_lining_SC_vol_perc', 'pkt_info_ClusterNum', 'pkt_info_Ngrid', 'pkt_info_Volume', 'pkt_info_Rinac(A) av', 'pkt_info_Rinac(A) mi', 'pkt_info_invRvolume(AA)', 'pkt_info_SITE_pocket_distance_min', 'pkt_info_num_adjacent_pockets', 'pkt_info_pocket_height', 'pkt_info_pocket_depth', 'pkt_info_metal_height', 'pkt_info_metal_depth', 'metal2_atom_name', 'metal2_serial', 'metal2_resName', 'metal2_seqID', 'metal3_atom_name', 'metal3_serial', 'metal3_resName', 'metal3_seqID', 'metal4_atom_name', 'metal4_serial', 'metal4_resName', 'metal4_seqID', 'Enzyme'],
       num_rows: 7844
    })
  })

which is a column oriented format that can be accessed directly, converted in to a pandas.DataFrame, or parquet format, e.g.

>>> MAHOMES_II.data.column('pdb')
>>> MAHOMES_II.to_pandas()
>>> MAHOMES_II.to_parquet("dataset.parquet")

Dataset Details

Dataset processing

We (Rosetta folks) merged the test split with the AlphaFold site annotations obtained from the MAHOMES GitHub repository using the struc_id identifier as the join key. This appended the corresponding Enzyme labels to each test entry while preserving all rows in the test dataset. We merged the training split with the MAHOMES-II site annotations obtained from the MAHOMES GitHub repository using struc_id as the join key. A left join was performed to append the corresponding Enzyme labels while preserving all entries in the training dataset.

Protein-metal interaction dataset

Chemical representation: PDB format with metal ions identified by residue ID (atomic symbol) as raw input. PDBs are then featurized for training/inference.

Protonation and tautomer states: No pH assumptions or charge assignments (seems that the PDBs are used as is, and metals are added to the proteins if the PDB was folded using AlphaFold.

Raw data → processed workflow

Original format: The original data is available either from the Protein DataBank or the AlphaFold Protein Structure Database. The structures are accessible via PDB or AF database ID, which are available in the MAHOMES_II_sites.csv AF_sites.csv

Storage location: Raw data are stored in the Protein Data Bank or AF structure database.

ML-oriented transformation

Feature extraction: Feature extraction is performed via the FeatureCalculations/ThirdPartyCalculations/runParty3Calc.sh script in the github repository. Feature calculations are performed using Rosetta, BLUUES, FindGeo, GHECOM, and pdb2pqr.

Representation/encoding: The representation/encoding is a concatenation of features (real values) calculated during the feature extraction step. The final representation is calculated in the MachineLearning/MAHOMES_II.py script in the github repository.

Encoding decisions: The encoding are one-hot encoded by the coordinating geometries of the protein structure.

Data validation and quality assessment

Missing data: Rosetta pocket measure did not detect surface pockets for 19% of the dataset, so GHECOM was used instead for detecting pockets. This method generated pockets for 99.5% of the dataset sites. Dataset sites that did not have a pocket were removed from the training data to improve dataset quality.

Training data labelling: Metal site labeling mislabeling was originally cross-validated with decision-tree ensemble machine learning model; then manually validated using literature, removing ambiguous entries.

Feature normalization: Normalization of features is performed using the scikit-learn QuantileTransformer method

Uses

This dataset is intended for training and evaluating machine learning models that classify protein-bound metal ions as enzymatic or non-enzymatic based on local structural and physicochemical features. It is well-suited for benchmarking binary classifiers on metalloprotein data, studying which physicochemical properties distinguish catalytic from non-catalytic metal sites, and supporting de novo metalloenzyme design by screening candidate structures for catalytic potential. The precomputed features (electrostatics, coordination geometry, Rosetta energy terms, and pocket descriptors) enable direct model training without requiring access to Rosetta, GHECOM, Bluues, or FindGeo.

Out-of-Scope Use

This dataset covers only metalloprotein sites coordinating Fe, Cu, Zn, Mn, Mg, Mo, Ni, and Co which is approximately 40% of all enzymes. It should not be used to make predictions about non-metal-dependent enzymatic activity, nor should it be treated as a general enzyme-versus-non-enzyme classifier for arbitrary protein sequences The dataset also excludes heme-containing sites, multi-chain metal binding sites, metal storage proteins, and NTPase sites, so models trained on it should not be expected to generalize to these classes.

Source Data

All data originates from protein crystal structures in the RCSB Protein Data Bank, filtered for resolution ≤3.5 Å and cross-referenced with the Mechanism and Catalytic Site Atlas (M-CSA) for enzymatic labeling. The dataset was constructed and curated by the Slusky Lab at the University of Kansas as part of the MAHOMES and MAHOMES II projects. Full details are described in Feehan, Franklin & Slusky (2021) Nature Communications 12:3712 and Feehan, Copeland, Franklin & Slusky (2023) Protein Science 32:e4626. Code and raw data are available at https://github.com/SluskyLab/MAHOMES-II.

Citation

@Article{MAHOMES-II, author = {Ryan Feehan, Matthew Copeland, Meghan W. Franklin, Joanna S. G. Slusky}, journal = {}, title = {MAHOMES II: A webserver for predicting if a metal binding site is enzymatic}, year = {2023}, doi = {} }