Our drug discovery business primarily revolves around hit identification, lead generation, and lead optimization to yield high-quality pre-clinical candidate molecules. Leveraging our integrated technology platform, we help transform the traditional methods for drug discovery and development and contribute to the pharmaceutical innovations in China and around the world.
Our innovative, integrated technology platform-based approach features an efficient, streamlined workflow with iterative steps including AI-powered molecule generation and comprehensive evaluation on drug-like properties such as selectivity, solubility, ADMET and synthesizability, prediction of molecular interactions using high-precision computational chemistry, and wet lab syntheses and validations. The diagram below illustrates the overall workflow underlying small molecule drug design.
We start from using our AI models to sample an ultra-large chemical space and generate tens of millions of drug-like molecules as the starting pool that our models predict to be suitable for the subsequent screening for the particular target at issue. As a broader and deeper search of the chemical space is conducted for each target as compared to traditional methods, our approach is more likely to yield quality candidate molecules for traditionally challenging targets.
Next, we deploy a combination of our AI models to perform a multi-objective optimization process where the AI models predict certain drug-like properties, such as solubility and ADMET features, some of which could only be assessed in later stage with traditional methods.
The optimal profile of a drug candidate represents an acceptable balance of drug-like properties such as potency, selectivity, solubility, bioavailability, half-life, permeability, drug-drug interaction potential, synthesizability, and toxicity, among others. We view drug design as a multi-objective optimization process because the several drug-like properties are often inversely correlated, meaning that optimizing one property often de-optimizes others. The inherent difficulties and uncertainties of achieving a balanced profile, the inability to assess certain critical drug-like properties and liabilities until in the later stage of development with traditional methods, and the limited sampling of chemical space often leads to suboptimal candidate molecules being advanced into subsequent development stages, which eventually results in costly late-stage failures. Therefore, we believe it is critical to identify potential failures early in the process of the drug R&D when they are relatively inexpensive, in order to increase the efficiency, reduce the overall cost and improve the success rate of the drug R&D projects.
The predictions yield a limited number of candidate molecules with a promising property profile. The algorithms and the prediction process are designed so that the resulting pool of molecules is small enough to be feasible for validation by the subsequent wet lab experimentation, while being large enough to ensure more than one candidate would meet the requirements for the next stage of R&D, leaving enough room for limited computation accuracy for the more challenging predictions.
Last, we perform wet lab experimentation to synthesize the pool of candidate molecules and conduct a variety of tests to validate their properties. We expect a vast majority of the standard synthesis and tests be carried out with high-throughput methods going forward to reduce costs, increase capacity, and improve repeatability. The wet lab validations also generate data on molecules which are used to train our in silico tools for better future insights.
Overall, we believe our approach enables discovery of high-quality molecules for potentially challenging targets at a more rapid pace and a larger scale, and with a higher likelihood of success compared to traditional methods.