I'm asking because it seems to vary widely. Some authors claim to have done "SAR" with 4-10 compounds, which seems silly because the "relationship" between descriptors/structural features and biological activity in these studies could be random and not statistically significant.
SAR studies with 20-35 compounds begin to demonstrate the complexity of the chemistry-bioactivity relationship and distinguish between well-correlated and poorly-correlated descriptors. SAR studies with 40-50 compounds allow for division into training and testing datasets and comparison of models but of course one must account for redundancy between the sets. I have heard talk of SAR studies with 100+ compounds, which sounds like a great expense but such data could be used in many interesting ways. I would love to do a survey of the literature and compare descriptors and sample sizes for predictive power across series. It would help to have some statistics.
(And it would be nice if more people included their raw data in their SI... If you have already published your results, what do you have to hide?? #RantoftheDay).
Number of compounds have not so much impact in a QSAR study,
The deciding factors are
* No. of descriptors (No. of Factors you are using in your QSAR study [i.e. no. of dependent variables and No. of independent variables]) used in study.
* Standard deviation near about 1 means your all compounds are similar so effect of variables or factors affecting will be similar, you model will be more reliable.
* so use variance & SD for selection of no. of compounds.
The rule of thumb for MLR based QSAR studies is 1:5 i.e. for every descriptor in the final model there should be five compounds at least in the data set.
The number of compounds taken is related to the number of descriptors chosen. However, the descriptor chosen also plays an important role. So depending on the significance of the descriptor, the number of compounds taken can vary.
The ideal ratio is 5:1.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size