I've been averaging Ki values from the PDSP Ki Database, and I can't help but to wonder: how I can determine p values for the binding affinities, so I can exclude anomalies in the data if need be?

  • $\begingroup$ What does your distribution of Ki look like? If it's reasonably normal, then convert to Z scores and pick an unacceptably low probability to discard (e.g., +/-3). $\endgroup$ – kmm Jun 23 '13 at 16:00
  • $\begingroup$ @kmm Ki values have to always be >0, yet normal distributions normally give a high probability for Ki values <0 and hence the normal distribution is something I'm wondering whether I should discard. It depends on the drug and receptor target too so I would like a nice and general answer if possible. I find that the kstest usually comes up for a p<0.1 that it fits a normal distribution so it might, but the negative values do concern me. $\endgroup$ – Josh Pinto Jun 23 '13 at 16:43
  • 1
    $\begingroup$ Can you add a histogram of your Ki values? Not all normal distributions have a mean of 0. $\endgroup$ – kmm Jun 23 '13 at 19:00

P-values reflect how well the data fits some background model. This background model should reflect some assumptions you make about the data. What assumptions do you think would be reasonable in this case? For example, do you assume high values are more likely to be incorrect?

Throwing away outliers like kmm suggested is probably not a good idea, because it means you will remove $K_i$ values which are very high or very low. It is not clear there is much biological justification for that, since we know some $K_i$s can be low and some can be high, but it doesn't tell us they are wrong.

The bottom line is that without making some assumption about what the values should look like, you can't identify what incorrect values look like.

| improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.