Edit | Radish peel
In the evolving field of drug discovery, traditional methods face significant challenges due to their low efficiency and high resource requirements.
Researchers at Model Medicines, an AI drug company, and the University of California, San Diego, have developed the GALILEO AI drug discovery platform and its core model, ChemPrint, with the goal of improving the efficiency of drug discovery.
To address the challenges of low hit rates and difficulty in exploring new chemical spaces, the platform employs adaptive molecular embedding and a rigorous model training environment to enhance prediction capabilities and navigate unknown molecular domains.
In the case for AXL and BRD4 oncology targets, ChemPrint achieved an in vitro hit rate of 45.5% and identified 20 novel acting compounds. These compounds exhibit great chemical novelty, with an average Tanimoto similarity score of 0.32 for their training set.
The study, titled "ChemPrint: An AI-Driven Framework for Enhanced Drug Discovery", was published on the bioRxiv preprint platform on March 27, 2024.
Traditional drug discovery methods face serious challenges such as high resource consumption and low efficiency, especially in the post-high-throughput screening (HTS) stage, which leads to the failure of a large number of potential drug candidates to be translated into clinical drugs.
In addition, existing AI models have limited ability to predict compound properties in unknown chemical spaces, and are susceptible to the constraints of chemical diversity in the training dataset.
Therefore, there is an urgent need to develop an AI-driven drug discovery framework that can effectively explore new chemical fields and accurately predict novel active compounds, so as to break through the traditional bottleneck and significantly improve the success rate of drug discovery.
To address these challenges, researchers from Model Medicines and the University of California, San Diego, proposed GALILEO, an artificial intelligence drug discovery platform, and its core component, ChemPrint, a zero-shot molecular geometric deep learning (Mol-GDL) model.
Illustration: GALILEO AI drug discovery platform workflow. (Source: Paper)
The GALILEO AI drug discovery platform is a comprehensive intelligent system designed to revolutionize the traditional drug discovery process through advanced artificial intelligence technology and dramatically improve the efficiency and success rate of screening new drug candidates.
The platform enhances the exploration of unknown chemical spaces by integrating large-scale high-throughput screening (HTS) data, leveraging adaptive molecular embeddings and a rigorous model training environment. The GALILEO platform maximizes the variability between the training and test sets through innovative data segmentation techniques, such as the t-SNE approach, to better predict and discover new compounds.
As a core component of the GALILEO platform, the ChemPrint model is an innovative AI model based on Mol-GDL. By learning the geometry and chemical properties of molecules, the model generates adaptable molecular embedding representations that can retain key chemical information.
ChemPrint's training and validation focus on real-world drug discovery scenarios rather than mere technical metrics has demonstrated superior predictive capabilities to accurately identify compounds with novel activity in unexplored chemical spaces.
Figure: t-SNE visualization of model latent embedding of AXL t-SNE segmentation data during the ChemPrint training cycle. (Source: Paper)
It achieved in vitro hits of up to 45.5% in cases targeting AXL and BRD4 oncology targets, a 75% improvement over the industry-published average, and identified 20 novel acting compounds. These compounds exhibit great chemical novelty, with an average Tanimoto similarity score of 0.32 for the training set, which is significantly different from known compounds and significantly exceeds traditional methods.
Illustration: Hits for in vitro validation of AXL and BRD4. (Source: Paper)
Researchers employ optimization techniques to reduce overfitting to create highly specialized adaptive molecular embeddings that confirm their effectiveness in prospective drug discovery studies. This advancement is critical to achieving clinical discovery.
All in all, GALILEO and ChemPrint effectively bridge the gap between the AI efficiency expected by academia and real-world treatment, drug discovery. The platform presented by the team solves some of the most difficult challenges in AI-driven drug discovery, offering a promising path in the pursuit of a more efficient and effective drug development process that will ultimately be powerful in saving lives.
Paper link: https://www.biorxiv.org/content/10.1101/2024.03.22.586314v1