Modeling Challenge - General Information

Updated: December 28, 2007 (when archived)

QSARWorld has created for you - our readers - a data modeling challenge. For this challenge we have decided to use a Human Oral Absorption data. You have to learn a model from the given training data and use that model for predicting the values for the test set. You send these predictions across to us and we declare the winner!

Rules & Regulations:

• An external test-set has been kept aside by us. Going by the usual Quantitative Structure-Activity Relationship (QSAR) modeling practice, it is 20% of the data. We have picked this set randomly by selecting proportionately from the activity column.


• Click here (not available now) for the "train" dataset in SDF file format.


• Click here (not available now) for the "external test" dataset in SDF file format.


• You may compute descriptors of your choice using any software. Those without ready access to a software tool may request for a fully functional free trial version of Sarchitect^TM by filling this form. Sarchitect^TM also allows one to build QSAR models. To learn more about Sarchitect^TM please click this link.

  For an overview video of Sarchitect^TM go to this link.



• You can also request us to send you a file (.txt,.csv) with the descriptors (and the end-point).


• You will need to send us your predictions for the "external test" dataset.


• You will need to send us a simple report on what you did - the descriptors and the modeling algorithm used, how you performed feature selection, the performance on the train set (train and cross-validation statistics) etc. for us to better understand your modeling approach. And of course, you have to send us your predictions for the external test set. Predictions below 0 or 100% should be threshold to 0 and 100, respectively.


• Performance on the external test set, in terms of the RMSE, will be the criterion for judging the modeling efforts. In case of a tie, preference will be given to simpler algorithm and lesser number of descriptors (not necessarily in that order). We will follow the philosophy outlined here for judgement in such a scenario.


• A book on Quantitative Structure-Activity Relationship modeling will be given as the top prize. Top 3 contributions will be listed at QSAR-World along with a report on their modeling efforts. Certificates will also be awarded by QSAR-World to the top 3 entries. Some other goodies may get bundled along with!


• Employees of Strand Life Sciences Pvt. Ltd., Bangalore, India and their families are not allowed to participate in this challenge.


• These rules and regulations are meant to be a guiding framework. As the competition progresses, based on the ensuing discussions, we all shall try to evolve our understanding and appreciation of the Quantitative Structure-Activity Relationship (QSAR) modeling methods.


• Please feel free to reach us at editor@qsarworld.com for questions, clarifications and comments on this challenge.

The poster about the Modeling Competition

The following poster is being sent out to various institutes announcing the modeling competition. If you want to receive a copy for your notice board write to us at editor@qsarworld.com providing us with your postal address.

Poster Design by: Suraj Vasisht & Shaillay Dogra