Predict in vivo parameter from in silico data.

Volume of distribution is calculated from following equation.
Vd = X / Cp0. X means dose, Cp0 means concentration of plasma at time 0. ( 1 compartment model)
This is the theoretical volume but important for PK/PD. If we could predict Vdss, we can predict dose of drug.
Recently I read the report written by scientist at Biogen.
The author used chemist friendly parameters for prediction because of the predict model tend to be black box.
In the report, the author used LogD of various PH range, molecular charge, and some donor acceptor parameters. I think these parameters are chemist friendly. 😉
They used to methods, PLS and Random Forest. RF showed better performance than PLS.
By the way, I was interested in the method of training, they used Leave-Class-Out(LCO) method instead of Leave-one-out(LOO).
LCO method is that remove one chemical series from training set and test model using the removed class. It indicates that test set is not similar to training set.
And surprisingly, the method works fine!
I think, key point of the report is the author used PhysChem parameters for prediction not used fingerprint. If molecular fingerprints were used to build the model, LCO method would not work (I think..).
Fortunately, the author provided dataset in supporting information. So reader who interested in the paper, reproduction or try to build another model is easy.
Prediction of vivo parameters is very attractive for me!



以下に詳細を記入するか、アイコンをクリックしてログインしてください。 ロゴ アカウントを使ってコメントしています。 ログアウト / 変更 )

Twitter 画像

Twitter アカウントを使ってコメントしています。 ログアウト / 変更 )

Facebook の写真

Facebook アカウントを使ってコメントしています。 ログアウト / 変更 )

Google+ フォト

Google+ アカウントを使ってコメントしています。 ログアウト / 変更 )

%s と連携中