Skip to content

extended fix to learning #611

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
carlocamilloni merged 3 commits into from
Feb 20, 2026
Merged

extended fix to learning #611

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
42500be
cleaning and extended the fix to the learning stage
carlocamilloni Feb 20, 2026
d308a71
extended the fix to make_mat
carlocamilloni Feb 20, 2026
2bc1de1
when learning only repulsive can be repulsive
carlocamilloni Feb 20, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
57 changes: 20 additions & 37 deletions src/multiego/ensemble.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -775,26 +775,21 @@ def init_LJ_datasets(meGO_ensemble, matrices, pairs14, exclusion_bonds14, args):
OO_mask = masking.create_linearized_mask(
type_ai_mapped.to_numpy(),
type_aj_mapped.to_numpy(),
[("O", "O"), ("OM", "OM"), ("O", "OM")],
[("O", "O"), ("O", "OM")],
symmetrize=True,
)

# hydrongen-oxygen attraction
HO_mask = masking.create_linearized_mask(
OMOM_mask = masking.create_linearized_mask(
type_ai_mapped.to_numpy(),
type_aj_mapped.to_numpy(),
[("H", "O"), ("H", "OM"), ("H", "OA")],
[("OM", "OM")],
symmetrize=True,
)

# oxygen-nitrogen repulsion (when not attractive)
ON_mask = masking.create_linearized_mask(
# hydrongen-oxygen attraction
HO_mask = masking.create_linearized_mask(
type_ai_mapped.to_numpy(),
type_aj_mapped.to_numpy(),
[
("O", "N"),
("OM", "N"),
],
[("H", "O"), ("H", "OM"), ("H", "OA")],
symmetrize=True,
)

Expand All @@ -805,11 +800,11 @@ def init_LJ_datasets(meGO_ensemble, matrices, pairs14, exclusion_bonds14, args):

# TODO
# here we should iterate over the pairs in type definition amd generate the repulsions
# NL-NZ repulsion
# NL-NL repulsion
NN_mask = masking.create_linearized_mask(
type_ai_mapped.to_numpy(),
type_aj_mapped.to_numpy(),
[("NL", "NL"), ("NZ", "NZ"), ("NL", "NZ")],
[("NL", "NL")],
symmetrize=True,
)

Expand All @@ -822,8 +817,8 @@ def init_LJ_datasets(meGO_ensemble, matrices, pairs14, exclusion_bonds14, args):
train_dataset.loc[OO_mask & ((train_dataset["bond_distance"] != 3) | (~train_dataset["same_chain"])), "rep"] = (
type_definitions.mg_OO_c12_rep
)
train_dataset.loc[ON_mask & ((train_dataset["bond_distance"] != 3) | (~train_dataset["same_chain"])), "rep"] = (
type_definitions.mg_ON_c12_rep
train_dataset.loc[OMOM_mask & ((train_dataset["bond_distance"] != 3) | (~train_dataset["same_chain"])), "rep"] = (
type_definitions.mg_OMOM_c12_rep
)
train_dataset.loc[HH_mask & ((train_dataset["bond_distance"] != 3) | (~train_dataset["same_chain"])), "rep"] = (
type_definitions.mg_HH_c12_rep
Expand All @@ -844,6 +839,7 @@ def init_LJ_datasets(meGO_ensemble, matrices, pairs14, exclusion_bonds14, args):
train_dataset["mg_sigma"] = pd.Series(pairwise_mg_sigma)
# special mg sigma cases:
train_dataset.loc[OO_mask, "mg_sigma"] = (type_definitions.mg_OO_c12_rep) ** (1 / 12) / 2 ** (1 / 6)
train_dataset.loc[OMOM_mask, "mg_sigma"] = (type_definitions.mg_OMOM_c12_rep) ** (1 / 12) / 2 ** (1 / 6)
train_dataset.loc[HH_mask, "mg_sigma"] = type_definitions.mg_HH_c12_rep ** (1 / 12) / 2 ** (1 / 6)
train_dataset.loc[NN_mask, "mg_sigma"] = (type_definitions.mg_NN_c12_rep) ** (1 / 12) / 2 ** (1 / 6)
train_dataset.loc[HO_mask, "mg_sigma"] = type_definitions.mg_HO_sigma
Expand Down Expand Up @@ -1074,6 +1070,7 @@ def set_sig_epsilon(meGO_LJ, parameters):
(meGO_LJ["probability"] <= meGO_LJ["limit_rc_att"] * np.maximum(meGO_LJ["rc_probability"], meGO_LJ["rc_threshold"]))
& (meGO_LJ["probability"] > meGO_LJ["md_threshold"])
& (meGO_LJ["rc_probability"] > meGO_LJ["md_threshold"])
& (meGO_LJ["epsilon_prior"] < 0.0)
)
meGO_LJ.loc[condition, "epsilon"] = (-meGO_LJ["rep"] * (meGO_LJ["distance"] / meGO_LJ["rc_distance"]) ** 12).clip(
lower=-20 * meGO_LJ["rep"], upper=-0.05 * meGO_LJ["rep"]
Expand Down Expand Up @@ -1612,12 +1609,8 @@ def make_pairs_exclusion_topology(meGO_ensemble, meGO_LJ_14, args):
] = type_definitions.mg_OO_c12_rep

df.loc[
(
(df["ai"].map(meGO_ensemble["sbtype_type_dict"]) == "OM")
)
& (
(df["aj"].map(meGO_ensemble["sbtype_type_dict"]) == "OM")
),
((df["ai"].map(meGO_ensemble["sbtype_type_dict"]) == "OM"))
& ((df["aj"].map(meGO_ensemble["sbtype_type_dict"]) == "OM")),
"c12",
] = type_definitions.mg_OMOM_c12_rep

Expand All @@ -1628,29 +1621,19 @@ def make_pairs_exclusion_topology(meGO_ensemble, meGO_LJ_14, args):
] = type_definitions.mg_HH_c12_rep

df.loc[
(
(df["ai"].map(meGO_ensemble["sbtype_type_dict"]) == "NL")
)
& (
(df["aj"].map(meGO_ensemble["sbtype_type_dict"]) == "NL")
),
((df["ai"].map(meGO_ensemble["sbtype_type_dict"]) == "NL"))
& ((df["aj"].map(meGO_ensemble["sbtype_type_dict"]) == "NL")),
"c12",
] = type_definitions.mg_NN_c12_rep

df.loc[
(
(
(df["ai"].map(meGO_ensemble["sbtype_type_dict"]) == "OM")
| (df["ai"].map(meGO_ensemble["sbtype_type_dict"]) == "O")
)
& ((df["aj"].map(meGO_ensemble["sbtype_type_dict"]) == "N"))
(df["ai"].map(meGO_ensemble["sbtype_type_dict"]) == "O")
& (df["aj"].map(meGO_ensemble["sbtype_type_dict"]) == "N")
)
| (
((df["ai"].map(meGO_ensemble["sbtype_type_dict"]) == "N"))
& (
(df["aj"].map(meGO_ensemble["sbtype_type_dict"]) == "OM")
| (df["aj"].map(meGO_ensemble["sbtype_type_dict"]) == "O")
)
(df["ai"].map(meGO_ensemble["sbtype_type_dict"]) == "N")
& (df["aj"].map(meGO_ensemble["sbtype_type_dict"]) == "O")
),
"c12",
] = type_definitions.mg_ON_c12_rep
Expand Down
11 changes: 0 additions & 11 deletions src/multiego/resources/type_definitions.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -204,23 +204,12 @@ def lj14_generator(df):
# List of atom type combinations for LJ14 pairs
atom_type_combinations = [
# Tuple of atom type combinations for LJ14 pairs
# ("backbone_carbonyl", "sidechain_cb", 0.275, 1.299682e-06, 1),
# ("backbone_oxygen", "sidechain_cb", 1, 1.5e-6, 0),
# ("ct_oxygen", "sidechain_cb", None, 5.0e-7, 0),
("backbone_nitrogen", "sidechain_cb", None, 5.0e-7, -1),
("first_backbone_nitrogen", "backbone_nitrogen", None, 4.0e-6, 1),
# ("backbone_nitrogen", "backbone_nitrogen", 0.343, None, 1),
# ("backbone_carbonyl", "backbone_carbonyl", 0.5, None, -1),
# ("sidechain_cgs", "backbone_carbonyl", 0.250, 1.2e-6, 0),
# ("sidechain_cgs", "backbone_nitrogen", 0.200, 5.5e-7, 0),
# ("sidechain_cgs", "first_backbone_nitrogen", 0.200, 5.5e-7, 0),
# ("sidechain_cds", "backbone_calpha", 0.100, 5e-7, 0),
]

# Special non-local interactions different from basic mg combination rules
# PROTEIN
polar_sbtype = ["O", "OA", "OM", "N", "NL", "NT", "NR", "NZ", "NE", "C", "S"]
hyd_sbtype = ["CH3", "CH2", "CH2r", "CH1", "CAH", "CAH2"]
special_non_local = [
{
"atomtypes": (["O"], ["O"]), # charged oxygen-oxygen repulsion
Expand Down
23 changes: 10 additions & 13 deletions tools/make_mat/make_mat.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -609,28 +609,25 @@ def main_routine(mol_i, mol_j, topology_mego, topology_ref, molecules_name, pref
OO_mask = masking.create_matrix_mask(
topology_df_i["mego_type"].to_numpy(),
topology_df_j["mego_type"].to_numpy(),
[("OM", "OM"), ("O", "O"), ("OM", "O")],
[("O", "O"), ("OM", "O")],
symmetrize=True,
)
HH_mask = masking.create_matrix_mask(
OMOM_mask = masking.create_matrix_mask(
topology_df_i["mego_type"].to_numpy(),
topology_df_j["mego_type"].to_numpy(),
[("H", "H")],
[("OM", "OM")],
symmetrize=True,
)
NN_mask = masking.create_matrix_mask(
HH_mask = masking.create_matrix_mask(
topology_df_i["mego_type"].to_numpy(),
topology_df_j["mego_type"].to_numpy(),
[("NL", "NL"), ("NZ", "NZ"), ("NL", "NZ")],
[("H", "H")],
symmetrize=True,
)
ON_mask = masking.create_matrix_mask(
NN_mask = masking.create_matrix_mask(
topology_df_i["mego_type"].to_numpy(),
topology_df_j["mego_type"].to_numpy(),
[
("O", "N"),
("OM", "N"),
],
[("NL", "NL")],
symmetrize=True,
)

Expand Down Expand Up @@ -665,10 +662,10 @@ def main_routine(mol_i, mol_j, topology_mego, topology_ref, molecules_name, pref

# define all cutoff using combination rule values and OO_mask
c12_cutoff = CUTOFF_FACTOR * np.power(np.where(OO_mask, type_definitions.mg_OO_c12_rep, c12_values), 1.0 / 12.0)
# apply OMOM correction
c12_cutoff = np.where(OMOM_mask, CUTOFF_FACTOR * np.power(type_definitions.mg_OMOM_c12_rep, 1.0 / 12.0), c12_cutoff)
# apply HH correction
c12_cutoff = np.where(HH_mask, CUTOFF_FACTOR * np.power(type_definitions.mg_HH_c12_rep, 1.0 / 12.0), c12_cutoff)
# apply ON correction
c12_cutoff = np.where(ON_mask, CUTOFF_FACTOR * np.power(type_definitions.mg_ON_c12_rep, 1.0 / 12.0), c12_cutoff)
# apply NN correction
c12_cutoff = np.where(NN_mask, CUTOFF_FACTOR * np.power(type_definitions.mg_NN_c12_rep, 1.0 / 12.0), c12_cutoff)

Expand All @@ -691,8 +688,8 @@ def main_routine(mol_i, mol_j, topology_mego, topology_ref, molecules_name, pref
1.0 / 12.0,
),
)
c12_cutoff = np.where(OMOM_mask, CUTOFF_FACTOR * np.power(type_definitions.mg_OMOM_c12_rep, 1.0 / 12.0), c12_cutoff)
c12_cutoff = np.where(HH_mask, CUTOFF_FACTOR * np.power(type_definitions.mg_HH_c12_rep, 1.0 / 12.0), c12_cutoff)
c12_cutoff = np.where(ON_mask, CUTOFF_FACTOR * np.power(type_definitions.mg_ON_c12_rep, 1.0 / 12.0), c12_cutoff)
c12_cutoff = np.where(NN_mask, CUTOFF_FACTOR * np.power(type_definitions.mg_NN_c12_rep, 1.0 / 12.0), c12_cutoff)

mismatched = topology_df_i.loc[topology_df_i["ref_type"].str[0] != topology_df_i["mego_name"].str[0]]
Expand Down
Loading