feat: add noncovalent-interactions

app.py

@@ -24,6 +24,7 @@
 from mlipaudit.dihedral_scan import DihedralScanBenchmark
 from mlipaudit.folding_stability import FoldingStabilityBenchmark
 from mlipaudit.io import load_benchmark_results_from_disk
+from mlipaudit.noncovalent_interactions import NoncovalentInteractionsBenchmark
 from mlipaudit.ring_planarity import RingPlanarityBenchmark
 from mlipaudit.small_molecule_minimization import (
     SmallMoleculeMinimizationBenchmark,
@@ -34,6 +35,7 @@
     conformer_selection_page,
     dihedral_scan_page,
     folding_stability_page,
+    noncovalent_interactions_page,
     ring_planarity_page,
     small_molecule_minimization_page,
     tautomers_page,
@@ -42,6 +44,7 @@
 BENCHMARKS: list[type[Benchmark]] = [
     ConformerSelectionBenchmark,
     DihedralScanBenchmark,
+    NoncovalentInteractionsBenchmark,
     TautomersBenchmark,
     RingPlanarityBenchmark,
     SmallMoleculeMinimizationBenchmark,
@@ -93,7 +96,14 @@ def _func():
     title="Tautomers",
     url_path="tautomers",
 )
-
+noncovalent_interactions = st.Page(
+    functools.partial(
+        noncovalent_interactions_page,
+        data_func=_data_func_from_key("noncovalent_interactions", data),
+    ),
+    title="Noncovalent Interactions",
+    url_path="noncovalent_interactions",
+)
 ring_planarity = st.Page(
     functools.partial(
         ring_planarity_page,
@@ -136,6 +146,7 @@ def _func():
         conformer_selection,
         dihedral_scan,
         tautomers,
+        noncovalent_interactions,
         ring_planarity,
         small_molecule_minimization,
         bond_length_distribution,

app_data/noncovalent_interactions/n_systems_per_subset.json

@@ -0,0 +1,46 @@
+{
+    "Dispersion: NobleGases": 140,
+    "Dispersion: PS": 103,
+    "Dispersion: HCNO": 79,
+    "Dispersion: Halogens": 94,
+    "Dispersion: Boron": 26,
+    "Hydrogen bonds: NH-O": 65,
+    "Hydrogen bonds: noHB": 113,
+    "Hydrogen bonds: CH-N": 20,
+    "Hydrogen bonds: OH-O": 60,
+    "Hydrogen bonds: NH-N": 53,
+    "Hydrogen bonds: OH-N": 45,
+    "Hydrogen bonds: CH-O": 19,
+    "Hydrogen bonds: XH-Cl": 32,
+    "Hydrogen bonds: XH-I": 19,
+    "Hydrogen bonds: XH-O": 51,
+    "Hydrogen bonds: XH-S": 54,
+    "Hydrogen bonds: XH-P": 52,
+    "Hydrogen bonds: XH-N": 34,
+    "Hydrogen bonds: XH-F": 41,
+    "Hydrogen bonds: XH-Br": 17,
+    "Ionic hydrogen bonds: NH(+)-O": 15,
+    "Ionic hydrogen bonds: OH-O(-)": 15,
+    "Ionic hydrogen bonds: OH-N(-)": 3,
+    "Ionic hydrogen bonds: CH-O(-)": 15,
+    "Ionic hydrogen bonds: NH(+)-N": 15,
+    "Ionic hydrogen bonds: NH-N(-)": 6,
+    "Ionic hydrogen bonds: CH-N(-)": 3,
+    "Ionic hydrogen bonds: NH-O(-)": 15,
+    "Ionic hydrogen bonds: NH-C(-)": 5,
+    "Ionic hydrogen bonds: NH(+)-C": 4,
+    "Ionic hydrogen bonds: OH-C(-)": 2,
+    "Ionic hydrogen bonds: OH(+)-O": 1,
+    "Ionic hydrogen bonds: CH-C(-)": 1,
+    "Repulsive contacts: NobleGases": 180,
+    "Repulsive contacts: Halogens": 235,
+    "Repulsive contacts: HCNO": 170,
+    "Repulsive contacts: PS": 154,
+    "Sigma hole: Cl": 29,
+    "Sigma hole: P": 33,
+    "Sigma hole: Br": 36,
+    "Sigma hole: S": 31,
+    "Sigma hole: Se": 44,
+    "Sigma hole: I": 42,
+    "Sigma hole: As": 35
+}

docs/source/api_reference/index.rst

@@ -11,6 +11,7 @@ API reference
     benchmark
     small_molecules/conformer_selection
     small_molecules/dihedral_scan
+    small_molecules/noncovalent_interactions
     small_molecules/tautomers
     small_molecules/ring_planarity
     small_molecules/minimization

docs/source/api_reference/small_molecules/noncovalent_interactions.rst

@@ -0,0 +1,22 @@
+.. _noncovalent_interactions_api:
+
+Noncovalent Interactions
+========================
+
+.. module:: mlipaudit.noncovalent_interactions.noncovalent_interactions
+
+.. autoclass:: NoncovalentInteractionsBenchmark
+
+    .. automethod:: __init__
+
+    .. automethod:: run_model
+
+    .. automethod:: analyze
+
+.. autoclass:: NoncovalentInteractionsResult
+
+.. autoclass:: NoncovalentInteractionsSystemResult
+
+.. autoclass:: NoncovalentInteractionsModelOutput
+
+.. autoclass:: NoncovalentInteractionsSystemModelOutput

docs/source/benchmarks/small_molecules/index.rst

@@ -12,6 +12,7 @@ sampling, stability, and interactions with other molecules.
 
     Conformer Selection <conformer_selection>
     Dihedral scan <dihedral_scan>
+    Noncovalent Interactions <noncovalent_interactions>
     Ring planarity <ring_planarity>
     Tautomers <tautomers>
     Minimization <minimization>

docs/source/benchmarks/small_molecules/noncovalent_interactions.rst

@@ -0,0 +1,74 @@
+.. _noncovalent_interactions:
+
+Non-covalent interactions
+=========================
+
+Purpose
+-------
+This benchmark tests if the **MLIP** can reproduce interaction energies of molecular complexes driven by non-covalent interactions.
+Non-covalent interactions are of highest importance for the structure and function of every biological molecule. This benchmark
+assesses a broad range of interaction types: London dispersion, hydrogen bonds, ionic hydrogen bonds, repulsive contacts and sigma
+hole interactions.
+
+
+Description
+-----------
+The benchmark runs energy inference on all structures of the distance scans of bi-molecular complexes in the dataset. The key
+metric is the **RMSE of the interaction energy**, which is the minimum of the energy well in the distance scan, relative to the
+energy of the dissociated complex - compared to the reference data. For repulsive contacts, the maximum of the energy profile is
+used instead. Some of the molecular complexes in the benchmark dataset contain exotic elements (see *dataset section*). In case that
+the **MLIP** has never seen an element of a molecular complex, this complex will be skipped in the benchmark.
+
+.. list-table::
+   :widths: 25 45
+   :header-rows: 0
+
+   * - .. figure:: img/butadiene_diazomethane.png
+          :width: 100%
+          :align: center
+          :figclass: align-center
+
+     - .. figure:: img/pes.png
+          :width: 100%
+          :align: center
+          :figclass: align-center
+
+Dataset
+-------
+This benchmark uses the datasets from the `NCI Atlas <http://www.nciatlas.org/>`_, with dissociation energy profiles.
+These datasets contain **QM** optimized geometries, along with **CCSD(T)/CBS** level interaction energies. The molecular complexes of
+these datasets contain typical organic small molecules, but also more exotic species and elements. Here is a summary of the
+datasets used in this benchmark:
+
+.. list-table:: NCI Atlas Datasets
+   :widths: 20 30 50
+   :header-rows: 1
+
+   * - Dataset Name
+     - Type of interaction
+     - Subsets
+   * - D442x10
+     - London dispersion
+     - Noble Gases, Boron, HCNO, Halogens
+   * - HB375x10
+     - Hydrogen bonds
+     - OH-N, OH-O, OH-C, NH-N, NH-O, …
+   * - HB300SPXx10
+     - Hydrogen bonds extended to S, P and halogens
+     - XH-S, XH-P, XH-Cl, XH-Br
+   * - IHB100x10
+     - Ionic hydrogen bonds
+     - O, N, C with cationic donors and anionic acceptors
+   * - R739x5
+     - Repulsive contacts
+     - HCNO, halogens, PS
+   * - SH250x10
+     - Sigma hole interactions
+     - P, S, Br, Cl, I
+
+Interpretation
+--------------
+The **RMSE** of the interaction energies should be **as low as possible**. This metric is likely to be very different for the different
+interaction types and data subsets. The **RMSE** in interaction error **should be compared per interaction type** and then in a more
+fine-grained visualization for the data subsets to identify areas of weakness for the **MLIP**. Within these areas of weakness,
+individual dissociation energy profiles can be visually inspected to see how they compare to the reference.

src/mlipaudit/main.py

@@ -26,6 +26,7 @@
 from mlipaudit.dihedral_scan import DihedralScanBenchmark
 from mlipaudit.folding_stability import FoldingStabilityBenchmark
 from mlipaudit.io import write_benchmark_results_to_disk
+from mlipaudit.noncovalent_interactions import NoncovalentInteractionsBenchmark
 from mlipaudit.ring_planarity import RingPlanarityBenchmark
 from mlipaudit.small_molecule_minimization import SmallMoleculeMinimizationBenchmark
 from mlipaudit.tautomers import TautomersBenchmark
@@ -35,6 +36,7 @@
 BENCHMARKS = [
     ConformerSelectionBenchmark,
     TautomersBenchmark,
+    NoncovalentInteractionsBenchmark,
     DihedralScanBenchmark,
     RingPlanarityBenchmark,
     SmallMoleculeMinimizationBenchmark,

src/mlipaudit/noncovalent_interactions/__init__.py

@@ -0,0 +1,18 @@
+# Copyright 2025 InstaDeep Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from mlipaudit.noncovalent_interactions.noncovalent_interactions import (
+    NoncovalentInteractionsBenchmark,
+    NoncovalentInteractionsResult,
+)