GithubHelp home page GithubHelp logo

mengtinghuang / fragment_based_retrosynthesis Goto Github PK

View Code? Open in Web Editor NEW

This project forked from knu-chem-lcbc/fragment_based_retrosynthesis

0.0 0.0 0.0 18.44 MB

Program and data for fragment-based retrosynthesis planning

Python 100.00%

fragment_based_retrosynthesis's Introduction

fragment_based_retrosynthesis

Datasets for fragment-based retrosynthesis planning

  1. PD1_RD1.txt.bz2
  • Single reactant reactions
  1. PD2_RD2.txt.bz2
  • Double reactant reactions
  1. PC6_RC6.txt.bz2
  • Mixed cases

Example entry from PD2_RD2.txt:

Tz Px rx dx E D L p b y v r h i o a t e dx D L v r o a t e - Tz p y i o a e


brief explanation of dataset curation

  1. We used the USPTO dataset recently mined by Lowe1. It contains 1,002,970 single product atom-mapped reactions. For the sake of simplicity, we restricted ourselves to single product reactions in this work. If it is considered that Lowe's dataset consists of 1,088,170 reactions in total with no duplicates, it is understood that limiting ourselves to single product reactions are not a preference but an obligation.
  • Number of reactions to start with : 1,002,970.
  1. We removed the reactions if the number of molecules at reactant side is greater than or equal to three.
  • The size of the dataset at this stage : 922,823.
  1. We also removed the duplicated reactions generated due to new representation.
  • The size of the dataset at this stage : 786,219.
  1. We cleaned internal twins, where product and reactant are identical on unimolecular reactions.
  • The size of the dataset at this stage : 780,471.

In this stage of the corpus; we have 186,206 unimolecular reactions, and 594,265 reactions where reactant side is composed of two molecules.

  1. We then examined the distribution of the reactions w.r.t total length of product and reaction pairs. For instance, The following reaction : [ C ix Wx I f u w o a e >> C I u w o a e - C ix Wx I f u w o a e ] has total length 27. We first set a maximum total length for a reaction as 100, and reduce the size of the dataset accordingly.
  • The size of the dataset at this stage : 429,889.
  1. For NMT applications, injection property is something desirable. In cases where a product maps into different reactants at different reactions, it is better having an injection from product domain to reactant domain. Therefore, we identified all such cases and applied a recipe to select only one of those reactions. We rioritized two reactants over a single reactant, and selected the one with minimum total lenght.
  • The size of the dataset at this stage : 352,546 (namely PC6_RC6.txt dataset which contains single and double reactant reactions)

License

Fragment-based retrosynthesis scheme by Juyong Lee and Umit Ucak is licensed under Attribution-NonCommercial-ShareAlike 4.0 International. To view a copy of this license, visit https://creativecommons.org/licenses/by-nc-sa/4.0

Footnotes

  1. Lowe, D.M.: Extraction of chemical structures and reactions from theliterature. PhD thesis, University of Cambridge (2012). โ†ฉ

fragment_based_retrosynthesis's People

Contributors

azpisruh avatar knu-chem-lcbc avatar

Recommend Projects

  • React photo React

    A declarative, efficient, and flexible JavaScript library for building user interfaces.

  • Vue.js photo Vue.js

    ๐Ÿ–– Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.

  • Typescript photo Typescript

    TypeScript is a superset of JavaScript that compiles to clean JavaScript output.

  • TensorFlow photo TensorFlow

    An Open Source Machine Learning Framework for Everyone

  • Django photo Django

    The Web framework for perfectionists with deadlines.

  • D3 photo D3

    Bring data to life with SVG, Canvas and HTML. ๐Ÿ“Š๐Ÿ“ˆ๐ŸŽ‰

Recommend Topics

  • javascript

    JavaScript (JS) is a lightweight interpreted programming language with first-class functions.

  • web

    Some thing interesting about web. New door for the world.

  • server

    A server is a program made to process requests and deliver data to clients.

  • Machine learning

    Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.

  • Game

    Some thing interesting about game, make everyone happy.

Recommend Org

  • Facebook photo Facebook

    We are working to build community through open source technology. NB: members must have two-factor auth.

  • Microsoft photo Microsoft

    Open source projects and samples from Microsoft.

  • Google photo Google

    Google โค๏ธ Open Source for everyone.

  • D3 photo D3

    Data-Driven Documents codes.