Benchmark and Interpretable MLIP, AtomBit-OMC, Released for Organic Molecular Crystals

AI editorPublished 2026.09.21Posted 2026.09.22 12:01 KST

OMC-bench and AtomBit-OMC were released, targeting structure relaxation, cocrystal reconstruction, polymorph energy ranking, and OOD force/stress prediction in organic molecular crystals (OMC). It was reported that the cocrystal structure reconstruction failure rate was the lowest and that polymorph ranking performance was the best, and that important covalent and noncovalent contacts were interpreted with a weight map. [1]

1/ AI utilization method: SE(3)-equivariant machine-learning interatomic potential and task-aligned benchmark
2/ Applied tasks: structure relaxation, cocrystal reconstruction, polymorph energy ranking, large-cell force and stress prediction for organic molecular crystals
3/ Reported results: the cocrystal structure reconstruction failure rate was the lowest, polymorph ranking performance was the best, and important contacts were presented in an interpretable way through a weight map
4/ OMC-bench and AtomBit-OMC is a study that jointly presents a benchmark and a machine-learning interatomic potential tailored to evaluating the tasks actually encountered in organic molecular crystals (organic molecular crystals).
5/ The benchmark includes OOD force and stress prediction in large cells, lattice dynamics, cocrystal (cocrystal) structure relaxation, and polymorph (polymorph) energy ranking.
6/ Based on this benchmark, the SE(3)-equivariant MLIP AtomBit-OMC was trained, and it was reported that the cocrystal structure reconstruction failure rate was the lowest and that the polymorph ranking metric was also the best.
7/ In addition, a gate-weight map was used to present chemically important covalent and noncovalent contacts in an interpretable way.
8/ From a materials perspective, this issue is important in that it combines practical benchmark design for structure prediction and stability evaluation of molecular crystals with MLIP development tailored to real crystal structures.
9/ However, the reported performance is limited to the benchmark and evaluation conditions, and it cannot be assumed to generalize directly to other molecular systems or other data distributions. [1]

How to apply this

To test an SE(3)-equivariant MLIP that handles both structure relaxation and energy ranking for organic molecular crystals.

1. Define the target molecular-crystal task and separate an evaluation set including cocrystal, polymorph, and distorted structures.
2. Generate labels for force, stress, and relative energy from reference calculations and include both large-cell and distorted cases.
3. Train an SE(3)-equivariant MLIP and validate it with metrics suitable for structure relaxation and energy ranking.
4. Check the weight or contribution interpretation to see whether important intermolecular contacts are physically reasonable.

Data needed: 3D structures of the target molecular crystal, candidate cocrystal and polymorph structures, and reference calculation data containing force and stress or energy labels
Checking the result: In a separate test set, check the force and stress errors, structure reconstruction success rate, and whether the polymorph ranking agrees with reference calculations.
What can go wrong: If the training data are biased toward a specific crystal system and deformation range, performance can drop sharply for new molecular combinations or large deformations.

이 가이드는 원문의 방법을 바탕으로 AI가 제안한 적용 방향입니다. 원문에 적힌 내용이 아니며 같은 결과가 재현된다는 보장도 아닙니다.

웹 자료를 바탕으로 AI가 자동 요약한 글입니다. 연구의 세부 조건과 결과는 연결된 원문을 확인해주세요.

Benchmark and Interpretable MLIP, AtomBit-OMC, Released for Organic Molecular Crystals

How to apply this

Related concepts and practice

Sources