Modeling a Crowdsourced Definition of Molecular Complexity

Abstract

This paper brings together the concepts of molecular complexity and crowdsourcing. An exercise was done at Merck where 386 chemists voted on the molecular complexity (on a scale of 1-5) of 2681 molecules taken from various sources: public, licensed, and in-house. The meanComplexity of a molecule is the average over all votes for that molecule. As long as enough votes are cast per molecule, we find meanComplexity is quite easy to model with QSAR methods using only a handful of physical descriptors (e.g., number of chiral centers, number of unique topological torsions, a Wiener index, etc.). The high level of self-consistency of the model (cross-validated R(2) ∼0.88) is remarkable given that our chemists do not agree with each other strongly about the complexity of any given molecule. Thus, the power of crowdsourcing is clearly demonstrated in this case. The meanComplexity appears to be correlated with at least one metric of synthetic complexity from the literature derived in a different way and is correlated with values of process mass intensity (PMI) from the literature and from in-house studies. Complexity can be used to differentiate between in-house programs and to follow a program over time.

The Synthesizability of Molecules Proposed by Generative Models

“…Figure S14: Correlation between the length of the first synthetic pathway found by ASKCOS and expert scores assigned by chemists in Sheridan et al. 17 A length of 0 indicates that the molecule can be found in our database of readily-purchasable compounds; a length of 11 indicates that no pathway was found with the fixed expansion settings (see Methods). Figure S15: Correlation between the length of the first synthetic pathway found by ASKCOS using all compound datasets and the SA Score 31 heuristic.…”

“…2 Current procedures for quantifying synthesizability are based on (1) structure complexity and similarity or (2) synthetic pathways. The structure-based approach usually involves constructing a heuristic definition based on domain expertise or chemical substructure diversity 14,15 or designing a model that can be fit to expert scores [16][17][18] or reaction data. 19,20 This kind of method is widely used due to its ease of implementation and low computational cost.…”

“…The most convincing metric might be a direct scoring from a group of experts on synthetic, medicinal chemistry, which has been used as a ground truth to train models against. [16][17][18][19] To have a group of experts that large enough to reach a non-biased and stable value is labor-intensive, hard to replicate, and not scalable. 21 The second, more nuanced approach to measuring synthesizability is to explicitly plan a synthetic pathway and assess its likelihood of experimental validity.…”

Computer-Assisted Retrosynthesis Based on Molecular Similarity

“…There have been many attempts to quantify synthetic accessibility, primarily involving heuristic scoring functions trained on subjective expert ratings. 56 Here, we use a success criterion that enables a more objective evaluation: that when given the products of reactions in the United States patent literature, the program recovers and ranks highly the recorded reactants without having seen that reaction previously.…”

mQC: A Heuristic Quality-Control Metric for High-Throughput Drug Combination Screening

“…In particular we based the development of the QC metric on a crowd-sourced assessment of response matrix quality. The use of crowd-sourced assessments of “quality” and other abstract descriptors has precedents14151617. For example, Lajiness et al 18.…”