Assessing the impact of a computational chemistry protocol [ACS 2024 Fall meeting poster]

Fu, Yuanxi; Zheng, Heng; Vandel, Ellie; Schneider, Jodi

Permalink

https://hdl.handle.net/2142/124768 Copy

Description

Title

Assessing the impact of a computational chemistry protocol [ACS 2024 Fall meeting poster]

Author(s)

• Fu, Yuanxi
• Zheng, Heng
• Vandel, Ellie
• Schneider, Jodi

Contributor(s)

• Sarol, M. Janina
• Sarraf, Ishita

Issue Date

2024-08-20

Keyword(s)

• Willoughby-Jansma-Hoye protocol
• unreliability propagation
• decision trees
• machine learning
• citation

Abstract

The Willoughby-Jansma-Hoye (WJH) protocol is a computational chemistry protocol that guides experimentalists in performing their own computational chemistry calculations to predict NMR spectra and assign the correct structures to organic molecules discovered. Its influence is attested by the number of citations (286 as of July 6, 2023, according to Web of Science and Scopus). In 2019, Neupane et al. discovered that one of the Python scripts supplied by the protocol contained a glitch: this script can produce erroneous NMR chemical shifts by mispairing the energy data with the NMR shielding tensor data in the Boltzmann weighting process if it is used on Linux-based platforms, resulting in an incorrect characterization of the structure. We designed three approaches to triage publications citing the WJH protocol based on their risk of further propagating unreliability from the code glitch: the base approach, the keyword-based approach (approach-KW), and the machine learning-based approach (approach-ML). We tested these approaches on an expert-annotated dataset of 284 publications citing the WJH protocol. The base approach uses the data that is easiest to retrieve. Approach-KW provides the most concrete rationales. Approach-ML has the best accuracy in triage results compared to experts’ annotations. We seek feedback from the chemistry community about the design of the three approaches.

Has Part

• https://doi.org/10.13012/B2IDB-4610831_V2
• https://doi.org/10.1145/3383583.3398514

Type of Resource

• still image
• text

Language

eng

Sponsor(s)/Grant Number(s)

• Alfred P. Sloan Foundation G-2020-12623
• Alfred P. Sloan Foundation G-2022-19409

Owning Collections