/sbmlcore

Collection of core classes and functions for structure-based machine learning to predict antimicrobial resistance

Primary LanguagePythonMIT LicenseMIT

Tests codecov PyPI version DOI

sbmlcore

Collection of core classes to help with building structure- and chemistry-based feature datasets to train machine learning models to predict antimicrobial resistance.

This is under active development and so is subject to change with no notice.

We will be making a series of jupyter-notebooks demonstrating how to use the classes available here.

Included features

Changes in Amino Acid Properties

  • Volume
  • Hydropathy scales: Kyte-Doolittle (paper) and WimleyWhite (paper)
  • Molecular weight
  • Isoelectric point

Secondary structure

Solvent accessible surface areas

Likelihood of changes in protein function

Effect of mutation on protein stability

  • DeepDDG: a more recent neural network that claims to outperform DUET, PopMusic etc. (paper and server). Can do all possible mutations in one job.

Structural distances

  • Distances between mutated residues and any atom/group of atoms of interest. Uses MDAnalysis (paper1 and paper2).

To potentially include at a later stage

  • Secondary structure: DSSP (do not anticipate much difference to STRIDE)
  • Protein stability:
    1. StabilityPredict. Online metapredictor, single amino acid at a time. Josh used in the pncA paper but had to contact them directly to run the entirity of PncA. (paper)
    2. DynaMUT. Also claims to outperform DUET etc. (paper). Can process a list of specified mutations in one job. (server)

PWF, 9 May 2023