New AI tool developed to revolutionize lead optimization in drug discovery

by

Lead optimization is a complex and time-consuming process in drug discovery that often relies on the expertise of medicinal chemists. However, this approach is subjective and can lead to uncertain outcomes and inefficiency. In order to address these challenges and streamline the process, researchers from the Shanghai Institute of Materia Medica (SIMM) of the Chinese Academy of Sciences have developed a groundbreaking artificial intelligence (AI) tool called pairwise binding comparison network (PBCNet).

The PBCNet utilizes a physics-informed graph attention mechanism to predict the relative binding affinity among congeneric ligands. By utilizing a pair of protein pocket-ligand complexes as input, this tool provides fast, precise, and user-friendly guidance for lead optimization in structure-based drug discovery.

To validate the performance of PBCNet, the researchers conducted extensive tests using two held-out sets provided by Schrodinger, Inc. and Merck KGaA. These sets consisted of over 460 ligands and 16 targets. The team incorporated transfer learning techniques, which involved pretraining the models on large-scale datasets and fine-tuning them for tasks with limited data. This approach significantly improved the models’ performance on the tasks.

The benchmarking results showed that the pretrained PBCNet outperformed other widely-used methods such as Schrodinger’s Glide, MM-GB/SA, and four recently reported deep learning models. Even with a small amount of fine-tuning data, PBCNet achieved comparable performance to Schrodinger’s FEP+, which is considered the gold standard in computational lead optimization.

Furthermore, the researchers tested the efficiency of PBCNet in real-world lead optimization scenarios. They used a benchmark consisting of nine chemical series and compared the order of model selection to the experimental order of synthesis. The results showed that incorporating PBCNet into the process accelerated lead optimization projects by approximately 473%, while reducing resource investment by an average of 30%.

This study highlights the immediate practical value of PBCNet in guiding lead optimization projects. In addition, an academic web service utilizing PBCNet to predict ligand binding affinity is freely available.

The development of PBCNet represents a significant advancement in the field of drug discovery. By incorporating domain-specific knowledge into its models, AI has the potential to revolutionize the lead optimization process and significantly accelerate the development of new drugs. The integration of physical and a priori knowledge into the modeling process behind PBCNet further demonstrates the power of AI in solving complex scientific problems.

*Note:

  1. Source: Coherent Market Insights, Public sources, Desk research
  2. We have leveraged AI tools to mine information and compile it