DeepSP: Deep Learning-Based Spatial Properties to Predict Monoclonal Antibody Stability

Abstract

Therapeutic antibody development, manufacturing, and administration face challenges due to high viscosities and aggregation tendencies often observed in highly concentrated antibody solutions. This poses a particular problem for subcutaneous administration, which requires low-volume and high-concentration formulations. The spatial charge map (SCM (mAbs, 8 (1) (2015), pp. 43-48)) and spatial aggregation propensity (SAP (PNAS. 2009; 106:11937-42) are two computational techniques proposed from previous studies to aid in predicting viscosity and aggregation, respectively. These methods rely on structural data derived from molecular dynamics (MD) simulations, which are known to be time-consuming and computationally demanding. DeepSCM (CSBJ. 2022, 20:2143-2152), a deep learning surrogate model to predict SCM scores in the entire variable region, was used to screen high-concentration antibody viscosity. DeepSCM is solely based on sequence information, which facilitates high throughput screening. This study further utilized a dataset of 20,530 antibody sequences to train a convolutional neural network deep learning surrogate model called Deep Spatial Properties (DeepSP). DeepSP directly predicts SAP and SCM scores in different domains of antibody variable regions based solely on their sequences without performing MD simulations. The linear correlation coefficient (R) between DeepSP scores and MD-derived scores for 30 properties achieved values between 0.76 and 0.96 with an average of 0.87 on the test set (N=2053). DeepSP was employed as features to build machine learning models to predict the aggregation rate of 21 antibodies. We observed remarkable results with R = 0.97 and a mean squared error (MSE) of 0.03 between the experimental and predicted aggregation rates, leave-one-out cross-validation (LOOCV) yielded R = 0.75 and MSE = 0.18, which is similar to the results obtained from the previous study using MD simulations. This result demonstrates that the DeepSP approach significantly reduces the computational time required compared to MD simulations. The DeepSP model enables the rapid generation of 30 structural properties that can also be used as features in other research to train machine learning models for predicting various antibody properties, such as viscosity, aggregation, or other properties that can influence their stability, using sequences only. The code and parameters are freely available at https://github.com/Lailabcode/DeepSP HighlightsO_LIDeep learning applied to develop a surrogate model (DeepSP) to rapidly predict 30 spatial properties of monoclonal antibodies that are usually calculated from MD simulations, using only sequences. C_LIO_LIThe DeepSP models achieved a linear correlation ranging between 0.76 and 0.96 with an average of 0.87, between the actual (MD simulation) and predicted score for all properties. C_LIO_LIDeepSP features were employed to build a model to predict aggregation rates of antibodies obtained from a previous study. A strong correlation of 0.97, and LOOCV correlation of 0.75 were achieved between the actual and predicted aggregation rates. C_LIO_LIDeepSP can be employed to generate antibody-specific features that can be used to train different machine learning models to predict antibody stability. C_LI