Artificial Intelligence Drug Discovery: Innovation in Pharmacy

Great minds always collaborate on problem-solving, and the pharmaceutical industry is the most prominent representation of this in relation to drug discovery. A critical factor in pharmaceutical research and development.

Innovation of this type takes planning and involves many moving parts. It requires extensive collaboration between chemistry, biology, toxicology, and pharmacokinetics to bring about one viable candidate to cure any disease.  

Fortunately, scientists are not thrown into the deep end to evaluate all these factors manually, as they used to in earlier decades.

The role of Artificial Intelligence in drug discovery is increasing, and according to McKinsey, incorporating AI capabilities into big data strategies has the potential to generate an annual value of up to $100 billion within the US healthcare system. 

This involves using predictive modelling and performing a thorough analysis of sensor data. 

What is Artificial Intelligence?  

The US Food and Drug Administration (FDA) defines AI as “the science and engineering of making intelligent machines”.

While Machine Learning is “an AI technique used to design and train software algorithms to learn from data.

It’s important to note that all machine learning (ML) techniques are considered AI techniques, but not all AI techniques involve ML. 

Standard AI workflows entail the following steps:

They also involve numerous algorithms and statistical modelling techniques that require in-depth knowledge from developers.  

How hard is it to create a new drug? 

The short answer is : it’s not easy. But recent technological advances have made pharmaceutical research and development more accessible and efficient.

To study each disease, scientists need to examine the receptors, enzymes, proteins, and genes associated with the disease. Once evaluated, the process of formulating a drug or treatment can begin. The diagram below shows the required drug discovery steps courtesy of Vatansever et al., 2021.  

The central aspect of drug discovery is to design a molecule that can reverse a disease by changing the activity of a target. This is why target identification is the first step in the process.

A good drug target needs to be relevant to the disease phenotype and suitable for therapeutic modulation (“druggable”).

Researchers also need to ensure the drug has a therapeutic benefit within an acceptable safety margin.

A scientist may realize that a drug is affecting the intended target but will still need to evaluate the mode of action to ensure patient safety using cell models, animal models, and, eventually, patient trials. 

How is AI being used in drug development? 

The analysis of large‐scale multidimensional biological data requires effective methods to produce accurate predictions for target identification.

AI is a robust technology used for analyzing the rapidly increasing multi-omics data in the identification of potential therapeutic targets.

In this section, we will discuss AI and drug discovery.  

Target prioritization

Nominating new drug targets and validating them can be expensive and time-consuming. Here, AI has been instrumental.

One ML method that has been applied to disease targets is SVM, which has historically been used to recognize handwriting, detect faces or identify a speaker.

Jeon et al. 2014 built an SVM classifier that uses features from various data types (DNA copy number, messenger RNA expression, mutation occurrence, and PPI) to prioritize drug targets specific to pancreatic, breast and ovarian cancers to identify and prioritize novel cancer drug targets.

The SVM algorithm is powerful, and they anticipate developing new kernel functions that can help in the discovery of new targets for heterogeneous cancers, such as triple-negative breast cancers (TNBCs) and soft tissue sarcomas (STS). 

Druggability

The likelihood of influencing a specific target with a small-molecule drug which depends on the target’s ability to bind small molecules based on its biophysical features.

AI drug discovery companies have addressed druggability by training models to estimate it using different properties. Including geometric, structural, and physicochemical features of drug‐binding and nondrug‐binding cavities on proteins.

This has helped researchers find that the most critical attributes to estimate druggability are the size and shape of the surface cavities of the protein.

For example: Costa et al., 2010 developed a decision tree-based meta-classifier by training on attributes such as network topological features, tissue expression profiles, and subcellular localization for each druggable and non-druggable gene.

It correctly identified 65% of known morbid genes with a precision of 66%. And correctly identified 78% of known druggable genes with a precision of 75%. This field still requires innovation. 

Absorption, Distribution, Metabolism, Excretion, and Toxicity (ADME‐T) predictions

ADME‐T properties are responsible for approximately half of all clinical failures, which is why it is essential to improve this process with AI applications.

Most prediction models attempt to build a direct relationship between molecular descriptors and ADME properties.

The introduction of capsule networks, a machine learning system that is a type of deep neural network (DNN) that can be used to model hierarchical relationships, has improved ADME‐T prediction.

For example, to predict the cardiotoxicity of drugs, Wang et al. 2019 developed two capsule network architectures: onvolution‐capsule network (Conv‐CapsNet) and restricted Boltzmann machine‐capsule network (RBM‐CapsNet) with 91.8% and 92.2% accuracy respectively. 

Need Pharmaceutical R&D Funding? Leyton can help

Pharmaceutical research and development is experiencing declining success rates and a stagnant pipeline.

Facilitating this type of innovation can be costly and time-consuming while posing great risk for pharmaceutical and AI drug discovery companies.

If you need funding for using Artificial Intelligence in drug discovery, Leyton has access to infinite grants that can help you cover the costs of this work.