Beyond tripeptides - two-step active machine learning for very large datasets

van Teijlingen, Alexander and Tuttle, Tell (2021) Beyond tripeptides - two-step active machine learning for very large datasets. Journal of Chemical Theory and Computation, 17 (5). pp. 3221-3232. ISSN 1549-9618 (https://doi.org/10.1021/acs.jctc.1c00159)

[thumbnail of Teijlingen-Tuttle-JCTC-2021-Beyond-tripeptides-two-step-active-machine-learning]
Preview
 Text. Filename: Teijlingen_Tuttle_JCTC_2021_Beyond_tripeptides_two_step_active_machine_learning.pdf
Final Published Version
License: Creative Commons Attribution 4.0 logo
Download (4MB)| Preview

Abstract

Self-assembling peptide nanostructures have been shown to be of great importance in nature and have presented many promising applications, for example, in medicine as drug-delivery vehicles, biosensors, and antivirals. Being very promising candidates for the growing field of bottom-up manufacture of functional nanomaterials, previous work (Frederix, et al. 2011 and 2015) has screened all possible amino acid combinations for di- and tripeptides in search of such materials. However, the enormous complexity and variety of linear combinations of the 20 amino acids make exhaustive simulation of all combinations of tetrapeptides and above infeasible. Therefore, we have developed an active machine-learning method (also known as "iterative learning" and "evolutionary search method") which leverages a lower-resolution data set encompassing the whole search space and a just-in-time high-resolution data set which further analyzes those target peptides selected by the lower-resolution model. This model uses newly generated data upon each iteration to improve both lower- and higher-resolution models in the search for ideal candidates. Curation of the lower-resolution data set is explored as a method to control the selected candidates, based on criteria such as log P. A major aim of this method is to produce the best results in the least computationally demanding way. This model has been developed to be broadly applicable to other search spaces with minor changes to the algorithm, allowing its use in other areas of research.

CORE (COnnecting REpositories)