When quoting this document, please refer to the following
DOI: 10.4230/DagSemProc.09081.3
URN: urn:nbn:de:0030-drops-20367
Go to the corresponding Portal

Walsh, Ian ; Vullo, Alessandro ; Pollastri, Gianluca

An adaptive model for learning molecular endpoints

09081.PollastriGianluca.Paper.2036.pdf (0.6 MB)


I will describe a recursive neural network that deals with undirected graphs, and its application to predicting property labels or activity values of small molecules.
The model is entirely general, in that it can process any undirected graph with a finite number of nodes by factorising it into a number of directed graphs with the same skeleton.
The model's only input in the applications I will present is the graph representing the chemical structure of the molecule. In spite of its simplicity, the model outperforms or matches the state of the art in three of the four tasks, and in the fourth is outperformed only by a method resorting to a very problem-specific feature.

BibTeX - Entry

  author =	{Walsh, Ian and Vullo, Alessandro and Pollastri, Gianluca},
  title =	{{An adaptive model for learning molecular endpoints}},
  booktitle =	{Similarity-based learning on structures},
  pages =	{1--16},
  series =	{Dagstuhl Seminar Proceedings (DagSemProc)},
  ISSN =	{1862-4405},
  year =	{2009},
  volume =	{9081},
  editor =	{Michael Biehl and Barbara Hammer and Sepp Hochreiter and Stefan C. Kremer and Thomas Villmann},
  publisher =	{Schloss Dagstuhl -- Leibniz-Zentrum f{\"u}r Informatik},
  address =	{Dagstuhl, Germany},
  URL =		{},
  URN =		{urn:nbn:de:0030-drops-20367},
  doi =		{10.4230/DagSemProc.09081.3},
  annote =	{Keywords: Recursive neural networks, qsar, qspr, small molecules}

Keywords: Recursive neural networks, qsar, qspr, small molecules
Collection: 09081 - Similarity-based learning on structures
Issue Date: 2009
Date of publication: 23.06.2009

DROPS-Home | Fulltext Search | Imprint | Privacy Published by LZI