Mobile Menu

An Overview of AI/ML for CNS Drug Discovery

Due to the complex nature of central nervous system diseases and the presence of the blood-brain barrier (among other factors), CNS drug discovery is particularly challenging. A recent study presented an overview of AI/ML-based approaches to meet challenges such as blood-brain barrier permeability in CNS drug discovery.

Central Nervous System (CNS) diseases

CNS diseases are a group of neurological disorders that have a significant economic and social impact. The development of new drugs for CNS diseases poses unique challenges when compared with other diseases, including the complexity of the brains anatomy and function, incomplete understanding of the biology of CNS diseases and the presence of the blood-brain barrier (BBB). Examples of CNS diseases include neurodevelopmental disorders, Parkinson’s disease and Alzheimer’s disease. The researchers behind this study present an overview of AI/ML-based approaches to meet the challenges associated with CNS drug discovery.

The importance of BBB permeability prediction in CNS drug discovery

The BBB is a biologic membrane that allows uptake of water, glucose, and essential amino acids as well as the efflux of small molecules and nonessential amino acids from the brain to the blood. While negligible penetration is desirable to minimise the brain side effects for peripheral drugs, high penetration is needed for CNS-active drugs to take their desired action. At present, researchers are required to carry out numerous experiments to assess whether a drug candidate is able to pass through the BBB, making the process expensive and time consuming. To improve the efficiency and success rate in CNS drug discovery, the BBB permeability of drug candidates needs to be assessed in the early drug discovery stages.

AI-based BBB permeability predictive models in drug discovery

In recent years, researchers have used AI-based predictive models to minimise the number of labour intensive and expensive BBB permeability experiments that need to be carried out in CNS drug discovery. Researchers have turned to various supervised learning approaches for constructing BBB permeability predictive models. These supervised learning approaches include support vector machine (SVM), Gaussian process, digital transformation (DT), K-nearest neighbour (KNN), linear discriminant analysis, consensus classifier and artificial neural networks (ANN). Researcgers originally developed these methods to process physical and chemical features including molecular weight, hydrophilicity, lipophilicity, topological polar surface area, acidic and basic atom numbers, hydrogen bond donors and acceptors, water-accessible volume, flexibility, van der Waals volume and ionisation potential.

Predictive capability of AI-based predictive models

The predictive capabilities of all the methods mentioned previously is limited to passive diffusional uptake and predominantly rely on few molecular descriptors. The issue then arises as many molecules, such as glucose and insulin, pass through the BBB via complex mechanisms that involve specific drug-transporter/drug-receptor interactions. Hence, such mechanisms are hard to describe via simple physiochemical features of compounds. Moreover, membrane transporters, such as the ATP-binding cassette, may make it difficult for drugs to achieve therapeutic drug concentrations. Although the primary role of these membrane transporters is limiting the brain entry of neurotoxins, they also limit the entry of many therapeutics and may contribute to CNS pharmacoresistance. Subsequently, prediction methods need to be capable of overcoming the limitations of physiochemical features and be able to address the multiple mechanisms associated with drugs that pass through the barrier.

SVM models for BBB permeability prediction in CNS drug discovery

To this end, Yuan et al. developed an SVM model that combined physiochemical properties and molecular fingerprints. The former is to account for passive diffusion, whereas the latter is to account for specific interactions such as uptake and efflux via binding proteins. When researchers compared this model with other SVM-based BBB permeability prediction methods it showed improved accuracy. This therefore demonstrates the importance of integrating physiochemical properties and fingerprints, taking into account the different methods of entry and efflux from the brain through the BBB.

Clinical trials of drug candidates often generate a large amount of phenotypic data in CNS, but the relationship between the CNS side-effects of drugs and their BBB permeation has not yet been captured. To try and bridge this knowledge gap, Gao et al.  developed a BBB permeability prediction SVM tool, which utilizes drug clinical phenotypes (drug side effects and drug indications). Although this study explored the BBB permeability prediction from a new angle by accounting for passive diffusion as well as putative contributions of active transport and other mechanisms, the accuracy of their SVM model still needs improvement. One reason for the reduced accuracy is that features based on physics and chemistry are different, hence the relation between drug side effects and therapeutic effects is more abstract. Therefore, traditional classification algorithms are not able to efficiently explore the relationship between data and results.

To overcome the issues with SVM approaches, researchers are using deep learning architectures, which have the ability to extract useful information from complex data structures with abstract relationships. Therefore, Miao et al. built a deep learning model to predict BBB permeability of drugs based on clinical features and this method achieved better performance than the other existing solutions.


This study outlined the issues associated with CNS drug discovery and outlined existing machine learning methods that researchers are using to predict BBB permeability in the early stages of drug discovery. The researchers state that for real progress to be made in the development of AI approaches, the BBB needs to be seen as a dynamic interface that might be affected by disease, and not just a physical barrier to drug delivery in the CNS.

Image credit: rawpixel – FreePik

Share this article