Poster Presentation HUPO 2019 - 18th Human Proteome Organization World Congress

DeepY: A deep learning model for biosimilarity evaluation of antibody drug using intact glycoproteins analysis by LC-MS (#424)

Geul Bang 1 , Ji won Lee 1 , Myung Jin Oh 2 , Hyun Joo An 2 , Heeyoun Hwang 1
  1. Korea Basic Science Institute, Cheongju, CHUNGBUK, South Korea
  2. Chungnam national university, Daejeon, South Korea

Heterogeneity and complexity of the glycosylation on biotherapeutics greatly depend on expression system, process conditions, and environment of cell culture of products. In order to evaluate biosimilarity of antibody drug such as Herceptin®, we have developed a deep learning model, DeepY, using intact glycoprotein analysis by LC-MS. Briefly, each antibody drug was independently analyzed to identify its intact glycoprotein composition. As a result, the list of identified intact glycoprotein compositions from each MS data was merged in the intact glycoprotein database, where a total of 34 intact glycoprotein compositions was identified from all three antibody drugs. Independently, the deconvoluted masses and their abundances of antibody drugs generated by MaxEnt were used as data sets such as training, validation, and test set, for development of a deep learning model using convolutional and fully connected neural network. The accuracy was approximately 90% at training, validation and test set. The DeepY could predict the biosimilarity and distinguish the low quality of mass spectra from all antibody drugs. We will further test the antibody drugs in batch to batch, expand to other original antibody drugs and their biosimilars.