when applying training and tests there are various methods people use depending on the size of your dataset. The percentage you use for each set is variable (you could use 75/25 split or 66/33 split). It is also advisable to run the training and test routine multiple times (maybe 100) with randomised sample orders to avoid bias in splitting. This also allows you to present standard deviation around your accuracy values.
One 1/3 test and 2/3 in training is good. Both training and test test should have a combination of Low, medium and highly active compounds.
Can we have toxic compound, with no given activity data, in our training and test data?
No. First you should find the QSAR formula using known test and training data. The formula should be able to predict the values for the test set more than 90 - 95% means then you can use this for compounds to predict the activity for which you donot have the experimental data.
no, actually i have experimental data but some of those compounds have toxic effects and not provided with their IC50...in one paper i ve read that in "activity column" compounds with no activity data should be represented by hyphen that is (-)...
- Badges
- Science topic
- Similar topics
- Bioinformatics
- Bioinformatics and Computational Biology