Guangdong Hec Technology Holding introduces the innovative HEC-PK model to reshape pharmacokinetics research.

In the wave of AI reshaping new drug development, Guangdong Hec Technology Holding Co., Ltd. has taken the lead in building an intelligent research and development system integrating "data-algorithm-platform". With more than ten years of research and development practice, covering multidimensional data including drug chemistry, biological activity, and pharmacokinetics, Guangdong Hec Technology Holding Co., Ltd. has used AI technology to construct a comprehensive rat PK curve prediction model 1.0 version (HEC-PK model), aiming to accurately predict the pharmacokinetic characteristics of drugs. The penetration of AI technology in the pharmaceutical field is extending from drug research and development to the entire industry chain. In the drug discovery stage, the introduction of large model technology has significantly improved the efficiency and novelty of new molecule design. As a pioneer in innovative drug research and development, Guangdong Hec Technology Holding Co., Ltd. is integrating AI technology into new drug development to assist in pharmacokinetic studies of drugs. Currently, based on data from the intravenous route (IV) of drug administration after data deduplication and organization, 1383 data points have been obtained and divided into training and testing sets in an 8:2 ratio, where the training set is used for training and the testing set is used to evaluate model performance. The HEC-PK model views small molecule compounds as graphs, and uses graph neural networks (GNN) as basic modules to extract feature information from molecular graphs, which can be used to predict rat PK curves. The model selection includes using GATConv as the message passing layer, and combining WeightAndSum Pooling + MaxPooling as the pooling layers. The current HEC-PK model has a depth of 2 layers, using 2 layers of GATConv for message passing, and pooling to extract information from the entire graph, predicting the PK curve of small molecule compounds using an MLP layer. The structure of the HEC-PK model is shown in the figure: Figure 1 Model Structure The company evaluates model performance by defining various performance metrics, where the logarithm average absolute error MAE (mean absolute error) and mfce (median fold change error) of the HEC-PK model in predicting concentration compared to actual concentration are close to 1, with a non-negative sample proportion of R2 exceeding 60% and an average non-negative sample R2 exceeding 0.6. In summary, the HEC-PK model performs admirably in predicting the pharmacokinetic PK curve of rats, demonstrating good predictive performance with small errors between predicted and actual PK curve values. HEC-PK has successfully built a comprehensive and accurate pharmacokinetic model, which will provide important theoretical support and application tools for drug development.