Ping An Insurance (Group) Company of China has reported that its researchers have come up with a deep learning framework for drug discovery.

A research team from Ping An Healthcare Technology Research Institute and Beijing's Tsinghua University developed the said framework; its findings were published in the peer-reviewed journal Briefings in Bioinformatics.

WHY IT MATTERS

The researchers created a new AI-driven framework for drug discovery called MPG that learns molecular representations from large volumes of unlabelled molecules. They also made their own graph neural networks (GNN) model called MolGNet for modelling molecular graphs.

Ping An said drug discovery can take between 10 to 15 years. AI technologies have been employed to speed up the process, particularly in molecule drug design, drug-drug interaction and drug-target interaction predictions. Yet, molecular designing remained a challenge given the dearth of labelled data for training datasets.

To this end, the research team worked with GNN technology, a model that can be pre-trained with unlabelled data instead of relying on labelled data.

In their research, the team crafted a self-supervised pre-training strategy named Pairwise Half-graph Discrimination. It found that after pre-training the MolGNet on 11 million unlabelled molecules, it captured "meaningful" patterns of molecules to produce an interpretable representation.

"The pre-trained MolGNet can be fine-tuned with just one additional output layer to create state-of-the-art models for a wide range of drug discovery tasks, including molecular properties prediction, drug-drug interaction and drug-target interaction, on 14 benchmark datasets," the researchers said in their study's abstract.

They also said their own GNN model has the potential to become an "advanced molecular encoder in the drug discovery pipeline".

Ping An Shionogi, Ping An's joint venture with Japanese pharmaceutical research firm Shionogi & Co. has utilised the AI framework for its research development of new drugs and drug repurposing.

THE LARGER TREND

In April, NVIDIA teamed up with AstraZeneca and the University of Florida to work on similar AI research projects to boost drug discovery.

They had revealed a new drug discovery model called MegaMoIBART for reaction prediction, molecular optimisation and de novo molecular generation.

Deployed on NVIDIA's Clara Discovery platform, the model will run on transformer neural networks technology, which does away with large labelled data sets and has been observed to help in accelerating drug discovery and research.