Script is an attempt to use principal component analysis (PCA) to identify key characteristics allowing to differ 'efficient' (working) reactions from 'inefficient' (not-working) ones.

For a random sample of chemical compounds, script employs retrosynthetic transforms developed at NU, to generate a space of all possible reactions. Then, it performs PCA for a predefined set of reaction descriptors.

Currently about a 100 of molecular descriptors are available. For each reaction a sum of values of a given molecular descriptor is calculated separately for its reactants and products. Compounds which H-depleted molecular graphs contains a single atom only (e.g. water 'O') are excluded, as many topological descriptors are not defined for them.

Each such a pair is treated as reaction descriptor corresponding to a given molecular descriptor.

To run the script type (currently all arguments are optional):

rxnpca.py --size <size> --seed <seed> --selection-type <type>

where

• size

  Sample size, i.e. number of chemicals for which reactions will be generated. Defaults to 1000.

• seed

  Pseudo-random number generator seed. Default to None, meaning that current system time will be used.

• type

  Selection method of reactions sharing the same product. Available choices are:

  • all: all available reactions will be selected,
  • random: a random pair of published and unpublished reactions will be selected for each product,
  • popular: a pair of published and unpublished reactions having the highest value of popularity index will be picked. If absent, default to all.