HamNet: Conformation-Guided Molecular Representation with Hamiltonian Neural Networks
This work addresses a bottleneck in molecular machine learning by improving 3D representations for applications in medical chemistry, though it is incremental relative to existing methods.
The paper tackles the challenge of incorporating 3D molecular conformations into deep learning representations by proposing HamNet, a Hamiltonian neural network that preserves conformations and achieves state-of-the-art performance on the MoleculeNet benchmark.
Well-designed molecular representations (fingerprints) are vital to combine medical chemistry and deep learning. Whereas incorporating 3D geometry of molecules (i.e. conformations) in their representations seems beneficial, current 3D algorithms are still in infancy. In this paper, we propose a novel molecular representation algorithm which preserves 3D conformations of molecules with a Molecular Hamiltonian Network (HamNet). In HamNet, implicit positions and momentums of atoms in a molecule interact in the Hamiltonian Engine following the discretized Hamiltonian equations. These implicit coordinations are supervised with real conformations with translation- & rotation-invariant losses, and further used as inputs to the Fingerprint Generator, a message-passing neural network. Experiments show that the Hamiltonian Engine can well preserve molecular conformations, and that the fingerprints generated by HamNet achieve state-of-the-art performances on MoleculeNet, a standard molecular machine learning benchmark.