Scalable Drug Property Prediction via Automated Machine Learning

The integration of artificial intelligence technologies into pharmaceutical research is crucial for gaining an early understanding of molecular properties, thereby facilitating successful drug design. Constructing a machine learning (ML) model however, requires knowledge spanning from data preprocessing and feature engineering to model fine-tuning, posing a challenge for chemists to effectively utilize ML tools for drug property predictions. This paper introduces a model training engine (MTE), which is a scalable automated ML pipeline that supports end-to-end \textit{in silico} drug property prediction. To accelerate the training process, a paralleled model fine-tuning scheme is developed for model optimization and selection, reducing the time complexity from $\mathcal{O}(n\times k)$ to $\mathcal{O}(n + k^2)$, where $ k >1$ and $k^2$ is much smaller than $n$. The MTE is benchmarked against five state-of-the-art models using twenty-two Therapeutic Data Commons ADMET datasets. The experimental results demonstrate the effectiveness and robustness of the MTE across diverse molecular data prediction tasks.