Strategy To Discover Diverse Optimal Molecules in the Small Molecule Universe

By Chetan Rupakheti, Aaron Virshup, Weitao Yang, David N. Beratan. 2015

86 citations recorded by the source corpus.

The small molecule universe (SMU) is defined as a set of over 1060 synthetically feasible organic molecules with molecular weight less than ∼500 Da. Exhaustive enumerations and evaluation of all SMU molecules for the purpose of discovering favorable structures is impossible. We take a stochastic approach and extend the ACSESS framework (Virshup et al. J. Am. Chem. Soc.2013, 135, 7296–73032354817710.1021/ja401184gPMC3670418) to develop diversity oriented molecular libraries that can generate a set of compounds that is representative of the small molecule universe and that also biases the library toward favorable physical property values. We show that the approach is efficient compared to exhaustive enumeration and to existing evolutionary algorithms for generating such libraries by testing in the NKp fitness landscape model and in the fully enumerated GDB-9 chemical universe containing 3 × 105 molecules.

Subfields: Computational Theory and Mathematics, Molecular Biology

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computational Theory and Mathematics

In Molecular Biology

Citations in the corpus

Grammar Variational Autoencoder (2017, background): “These databases are too large to search via exhaustive enumeration, and require novel stochastic search algorithms tailored to the domain (Virshup et al., 2013; Rupakheti et al., 2015).”
Chemical Space Mimicry for Drug Discovery (2017, methodology): “In particular, Vishrup and Rupakheti 1,2 described an iterative method to enumerate compounds over all of chemical space in a way that maximizes structural diversity and demonstrated the potential of this approach toward drug design applications.”
Self-referencing embedded strings (SELFIES): A 100% robust molecular string representation (2019, background): “challenging before. Concrete examples are evolutionary strategies, which can be applied as a robust alternative to Q-learning [14], and as direct generative models for the de novo design of molecules [15, 16, 17, 18]. 2 Related Work A large amount of effort has been put into the construction of repr”
SMILES-BERT: Large Scale Unsupervised Pre-Training for Molecular Property Prediction (2019, methodology): “Another stream of traditional fingerprint methods are based on the biological experiments and the expertise knowledge and experience, e.g., [22, 28].”