Autoencoding Undirected Molecular Graphs With Neural Networks
Authors:
Jeppe Johan Waarkjær Olsen,
Peter Ebert Christensen,
Martin Hangaard Hansen,
Alexander Rosenberg Johansen
Abstract:
Discrete structure rules for validating molecular structures are usually limited to fulfillment of the octet rule or similar simple deterministic heuristics. We propose a model, inspired by language modeling from natural language processing, with the ability to learn from a collection of undirected molecular graphs, enabling fitting of any underlying structure rule present in the collection. We in…
▽ More
Discrete structure rules for validating molecular structures are usually limited to fulfillment of the octet rule or similar simple deterministic heuristics. We propose a model, inspired by language modeling from natural language processing, with the ability to learn from a collection of undirected molecular graphs, enabling fitting of any underlying structure rule present in the collection. We introduce an adaption to the popular Transformer model, which can learn relationships between atoms and bonds. To our knowledge, the Transformer adaption is the first model that is trained to solve the unsupervised task of recovering partially observed molecules. In this work, we assess how different degrees of information impact performance w.r.t. to fitting the QM9 dataset, which conforms to the octet rule, and to fitting the ZINC dataset, which contains hypervalent molecules and ions requiring the model to learn a more complex structure rule. More specifically, we test a full discrete graph with bond order information, a full discrete graph with only connectivity, a bag-of-neighbors, a bag-of-atoms, and a count-based unigram statistics. These results provide encouraging evidence that neural networks, even when only connectivity is available, can learn arbitrary molecular structure rules specific to a dataset, as the Transformer adaption surpasses a strong octet rule baseline on the ZINC dataset.
△ Less
Submitted 21 March, 2020; v1 submitted 26 November, 2019;
originally announced January 2020.
An Atomistic Machine Learning Package for Surface Science and Catalysis
Authors:
Martin Hangaard Hansen,
José A. Garrido Torres,
Paul C. Jennings,
Ziyun Wang,
Jacob R. Boes,
Osman G. Mamun,
Thomas Bligaard
Abstract:
We present work flows and a software module for machine learning model building in surface science and heterogeneous catalysis. This includes fingerprinting atomic structures from 3D structure and/or connectivity information, it includes descriptor selection methods and benchmarks, and it includes active learning frameworks for atomic structure optimization, acceleration of screening studies and f…
▽ More
We present work flows and a software module for machine learning model building in surface science and heterogeneous catalysis. This includes fingerprinting atomic structures from 3D structure and/or connectivity information, it includes descriptor selection methods and benchmarks, and it includes active learning frameworks for atomic structure optimization, acceleration of screening studies and for exploration of the structure space of nano particles, which are all atomic structure problems relevant for surface science and heterogeneous catalysis. Our overall goal is to provide a repository to ease machine learning model building for catalysis, to advance the models beyond the chemical intuition of the user and to increase autonomy for exploration of chemical space.
△ Less
Submitted 1 April, 2019;
originally announced April 2019.
