search

openvax / mhcflurry

Peptide-MHC I binding affinity prediction
http://openvax.github.io/mhcflurry/
Apache License 2.0
193 stars 58 forks source link

what are the units of your predictions? #52

Closed flamholz closed 8 years ago

flamholz commented 8 years ago

Is the absolute value meaningful? Is it a -log(KD) or something? Can this be documented in the README?

Avi

timodonnell commented 8 years ago

If you just call predict, the values returned are affinities (KD) in nM. Values < 500 are often considered "binders".

Internally, we actually predict a rescaled quantity: 1 - log(a)/log(50000) where a is the nM affinity. But users shouldn't have to interact with this rescaled quantity.

Does that answer your question?

timodonnell commented 8 years ago

Added a note to the readme

flamholz commented 8 years ago

Yup! Thanks for fast answer!

But it should be documented in the readme, since you give an example but don't give units and it may not be apparent from that one example whether the (1) the prediction is relative or absolute (i.e. whether it has meaning in the absence of a second prediction) and (2) whether it is in uM, nM, fM units.

On Sat, Sep 17, 2016 at 7:41 PM timodonnell notifications@github.com wrote:

If you just call predict, the values returned are affinities (KD) in nM. Values < 500 are often considered "binders".

Internally, we actually predict a rescaled quantity: 1 - log(a)/log(50000) where a is the nM affinity. But users shouldn't have to interact with this rescaled quantity.

Does that answer your question?

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/hammerlab/mhcflurry/issues/52#issuecomment-247814791, or mute the thread https://github.com/notifications/unsubscribe-auth/ACg2geMlg2D8Kz9td-UvZEfloeFaXS7Vks5qrHqbgaJpZM4J_vlX .

flamholz commented 8 years ago

I am satisfied and educated!