Year: 2019
Type: dissertation
Abstract: Natural products have been the single most productive source of lead compounds for the modern drug development. In traditional drug discovery from natural products, concentrated extracts prepared from, e.g., plant samples were screened to determine their bioactivity. These extracts are complicated mixtures. Thus, a signal from the screening assay may be confounded, e.g., by synergistic effects of several compounds. However, isolating each compound from the extract prior to the screening would be inefficient when a large number of samples are screened. Structures of compounds in a natural product sample are unknown in advance. Analytical methods, such as tandem mass spectrometry (MS/MS), are used to identify the constituents of the samples in almost every stage of the drug discovery process from natural products. We argue that predicting bioactivities based on MS/MS spectra could be used to prioritize the most promising samples for further experimental testing. We introduce two machine learning pipelines to predict bioactivities from MS/MS spectra. First, we predict bioactivities directly from MS/MS spectra. Second, we train a model to identify an unknown compound based on its MS/MS spectrum and another model to predict bioactivities given a compound with known structure. In the testing phase, structure predicted from an MS/MS spectrum is used to predict bioactivities. In the first pipeline, only drugs which have both MS/MS spectrum and bioactivities available can be used in the training. However, the overlap of MS/MS and bioactivity datasets is limited. Advantage of the second approach is its ability to use drugs which have either MS/MS spectrum or bioactivities available in the training. We show that the second approach results in more accurate predictions compared to the first approach. Additionally, we show that we can a build predictive model even in case there is no overlap of the drugs in the MS/MS and the bioactitivity datasets which is not possible with the first approach.
Source: Aaltodoc (Aalto University)
Author: Jägerroos, Vilma
Language: English
Subfield: Drug Discovery
Domain: Life Sciences
Open Access status: green
OpenAlex ID: W7112966743