Dataset’s chemical diversity limits the generalizability of machine learning predictions

Glavatskikh, Marta; Leguy, Jules; Hunault, Gilles; Cauchy, Thomas; Mota, Benoit Da

doi:10.1186/s13321-019-0391-2

Cited by 94 publications

(145 citation statements)

References 53 publications

(57 reference statements)

Supporting

Mentioning

121

Contrasting

Order By: Relevance

“…Creating a chemical space that reflects the similarity in the physical properties as well as molecular structures will drastically improve the efficiency of molecular design. And a wide diversity of molecular property in the dataset is important to generalize prediction [2] . In this paper, we introduce a model to embed molecules into a latent space using deep learning to reproduce the distance between the properties of chemicals based on their molecular structures.…”

Section: Figurementioning

confidence: 99%

Embedding of Molecular Structure Using Molecular Hypergraph Variational Autoencoder with Metric Learning

Koge

Ono

Huang

et al. 2020

Molecular Informatics

View full text Add to dashboard Cite

Deep learning approaches are widely used to search molecular structures for a candidate drug/material. The basic approach in drug/material candidate structure discovery is to embed a relationship that holds between a molecular structure and the physical property into a low‐dimensional vector space (chemical space) and search for a candidate molecular structure in that space based on a desired physical property value. Deep learning simplifies the structure search by efficiently modeling the structure of the chemical space with greater detail and lower dimensions than the original input space. In our research, we propose an effective method for molecular embedding learning that combines variational autoencoders (VAEs) and metric learning using any physical property. Our method enables molecular structures and physical properties to be embedded locally and continuously into VAEs’ latent space while maintaining the consistency of the relationship between the structural features and the physical properties of molecules to yield better predictions.

show abstract

Section: Figurementioning

confidence: 99%

Embedding of Molecular Structure Using Molecular Hypergraph Variational Autoencoder with Metric Learning

Koge

Ono

Huang

et al. 2020

Molecular Informatics

View full text Add to dashboard Cite

show abstract

“…In a previous article we studied the QM9 and PC9 datasets that together encompass more than 200k different molecular calculations with up to 9 heavy atoms of C, N, O and F types [ 31 ]. Initially computed with different methods, we relaunched them using our BOINC collaborative computing project, called QuChemPedIA@home, in order to have a homogeneous and clean dataset.…”

Section: Resultsmentioning

confidence: 99%

“…One could hope for a fast evaluation of quantum mechanics properties thanks to machine learning predictions to limit the cost of computation. However, we have demonstrated that the currently available datasets of molecular quantum chemistry results, like QM9 and PC9, are not diverse enough to train a general predictor [ 31 ]. It is clear that solving this issue in the future would significantly accelerate the generation of molecules with such objectives.…”

Section: Resultsmentioning

confidence: 99%

See 1 more Smart Citation

EvoMol: a flexible and interpretable evolutionary algorithm for unbiased de novo molecular generation

Leguy¹,

Cauchy

Glavatskikh

et al. 2020

J Cheminform

Self Cite

View full text Add to dashboard Cite

The objective of this work is to design a molecular generator capable of exploring known as well as unfamiliar areas of the chemical space. Our method must be flexible to adapt to very different problems. Therefore, it has to be able to work with or without the influence of prior data and knowledge. Moreover, regardless of the success, it should be as interpretable as possible to allow for diagnosis and improvement. We propose here a new open source generation method using an evolutionary algorithm to sequentially build molecular graphs. It is independent of starting data and can generate totally unseen compounds. To be able to search a large part of the chemical space, we define an original set of 7 generic mutations close to the atomic level. Our method achieves excellent performances and even records on the QED, penalised logP, SAscore, CLscore as well as the set of goal-directed functions defined in GuacaMol. To demonstrate its flexibility, we tackle a very different objective issued from the organic molecular materials domain. We show that EvoMol can generate sets of optimised molecules having high energy HOMO or low energy LUMO, starting only from methane. We can also set constraints on a synthesizability score and structural features. Finally, the interpretability of EvoMol allows for the visualisation of its exploration process as a chemically relevant tree.

show abstract

“…including ANI-1, 14,15 PC-9, 69,70 and ISO-17, [71][72][73] and more are continuously added. Datasets are ingested into a common format of structured metadata and are available for download in text format, or in structured HDF5 format.…”

Section: Machine Learning Datasets Web Appmentioning

confidence: 99%

The MolSSI QCArchive Project: An Open-Source Platform to Compute, Organize, and Share Quantum Chemistry Data

Smith

Altarawy²,

Burns³

et al. 2020

Preprint

View full text Add to dashboard Cite

<div>The Molecular Sciences Software Institute's (MolSSI) Quantum Chemistry Archive (QCArchive) project is an umbrella name that covers both a central server hosted by MolSSI for community data and the Python-based software infrastructure that powers automated computation and storage of quantum chemistry results.</div><div>The MolSSI-hosted central server provides the computational molecular sciences community a location to freely access tens of millions of quantum chemistry computations for machine learning, methodology assessment, force-field fitting, and more through a Python interface.</div><div>Facile, user-friendly mining of the centrally archived quantum chemical data also can be achieved through web applications found at https://qcarchive.molssi.org.</div><div>The software infrastructure can be used as a standalone platform to compute, structure, and distribute hundreds of millions of quantum chemistry computations for individuals or groups of researchers at any scale.</div><div>The QCArchive Infrastructure is open-source (BSD-3C), code repositories can be found at https://github.com/MolSSI, and releases can be downloaded via PyPI and Conda.</div>

show abstract

Dataset’s chemical diversity limits the generalizability of machine learning predictions

Cited by 94 publications

References 53 publications

Embedding of Molecular Structure Using Molecular Hypergraph Variational Autoencoder with Metric Learning

Embedding of Molecular Structure Using Molecular Hypergraph Variational Autoencoder with Metric Learning

EvoMol: a flexible and interpretable evolutionary algorithm for unbiased de novo molecular generation

The MolSSI QCArchive Project: An Open-Source Platform to Compute, Organize, and Share Quantum Chemistry Data

Contact Info

Product

Resources

About