For full text search please use the '?' prefix. e.g. ? Onboarding

A Sample Centric And Knowledge Driven Computational Framework For Natural Products Drug Discovery

A Sample-Centric and Knowledge-Driven Computational Framework for Natural Products Drug Discovery

Terms

  • NP: Natural products
  • NE: Natural Extact(s)
  • ENPKG: Experimental Natural Products Knowledge Graph
  • UPLC-HRMS: Ultrahigh performance liquid chromatography coupled with tandem high-resolution mass spectrometry.
  • RT: Retention time
  • FBMN: (Feature-based molecular networking) technique for organizing fragmentation spectra into similar clusters to analyze spectra.
  • RDF: Research description framework (subject-predicate-object)
  • FAIR: (Findability, accessibility, interoperability, reusability)
  • MN: Molecular Network
  • DOI: Digital Object Identifier
  • USI: Universal Spectrum Identifier

1. intro

the complexity of diverse compounds with NEs. So determining NEs precisely is an issue. Currently this is done with UPLC-HRMS. Then FBMN is used for analysis. This type of analysis requires a lot of resources for larger datasets. This approach is data set-centric, but the authors of this paper are arguing for a sample-centric approach. In their approach each sample is considered invidivually and then compiled into a knowledge graph of all samples under consideration. They implement this as the ENPKG.

In their approach they are able to utilize RDF and therefore SPARQL for querying.

2. Results

NPs are characterized by their metadata. such as:

  • taxonomy
  • part (of plant) studied
  • extract type
  • bioactivity
  • (and more) The LC-HRMS2 data can be gathered in two ways: 1) directly generated by researchers 2) reusing published data sets the data either way is processed and used to do feature detection. Then the files are uploaded to the MassIVE repo, and can therefore recieve a DOI and each MSMS spectrum gets a USI.

Data is processed by: 1) taxonomical standardization 2) local implementation of FBMN 3) chemical structure annotation using ISDB-LOTUS 4) chemical structure and chemical class annotationusing SIRIUS' FingerID and CANOPUS

Then the data is converted into RDF and a KG is constructed from the triples. Then you can run SPARQL queries over the data.

PhD Connection Questions

Problem being solved

Batch effects caused by recording data in different batches, need for recomputation of novel samples into already analyzed data. This is due to being set-centric and divides things up into different separate projects that are isolated (siloed).

Data inputs

list of LC-MS features: m/z, RT, area/intensity, and MS/MS spectrum.

Assumptions about input data

Large number of samples in the dataset.

Transformation pipeline

Each sample is used individually for taxonomic metadata standardization, feature detection, structural annotation, and FBMN (feature based molecular networking).

Semantic model / ontologies

the data is transformed into RDF (resource description framework) triple data (subject - predicate - object) and placed into a knowledge graph. Data could look like: molecule A found in Species X.

MS data starts by gaining a DOI (from MassIVE) and USI (universal spectrum identifier), and metadata includes:

  • LC-HRMS2 filename(s)
  • sample id
  • sample type
  • sample source id
  • sample's source taxonomical denomination Each sample is organized as a directory.