- Global Pharma News & Resources

Artificial Intelligence in Chemical Drug Development and Molecular Target Identification


Artificial Intelligence (AI) is the simulation of human intelligence processes by computers. The process includes obtaining information, formulating rules for using information, drawing approximate or definite conclusions, and self-correction.
  • Author Name: Lisa George
Editor: Lisa George Last Updated: 28-Jan-2021

Artificial Intelligence (AI) is the simulation of human intelligence processes by computers. The process includes obtaining information, formulating rules for using information, drawing approximate or definite conclusions, and self-correction.

Artificial intelligence, machine learning and deep learning

The method of using AI in discovering drug-like small molecules involves the use of chemical space. By enumerating possible organic molecules by calculation, the chemical space provides a stage for the identification of new and high-quality molecules. In addition, machine learning technology and predictive model software can also help identify target-specific virtual molecules and the association of molecules with their respective targets, while optimizing safety and efficacy attributes.

AI systems can reduce the wastage rate and R&D expenditure by reducing the number of synthetic compounds that are later tested in in vitro or in vivo systems. Proven AI technology can be used to improve the success rate of drug development, and the AI technology under development must be verified before being applied to the drug development process. The most critical part of the drug development process is the synthesis of selected molecules. Therefore, AI is of great value because it can prioritize molecules based on the ease of synthesis or develop tools that are effective for the best synthetic route.

The significance of AI in synthetic drug compounds

Speaking of AI-assisted new chemical drug development, machine learning (ML), a subfield of artificial intelligence, should be mentioned. In 1959, one of the pioneers of machine learning, Arthur Samuel, used machine learning as a research field, enabling computers to learn without explicit programming.

Machine learning is divided into supervised learning, unsupervised learning and reinforcement learning. Supervised learning includes classification and regression methods, in which predictive models are developed based on data from input and output sources. The output of supervised machine learning requires disease diagnosis under subgroup classification, and ADMET prediction under drug efficacy and subgroup regression. Unsupervised learning includes methods of clustering and feature search by grouping and interpreting data based only on input data. Through unsupervised machine learning, output such as discovering disease subtypes from clusters and discovering disease targets from feature discovery methods can be achieved. Reinforcement learning is mainly driven by decision-making and its execution in a given environment to maximize its performance. The output of this type of ML includes de novo drug design in decision-making and experimental design in implementation. Both can be achieved through modeling and quantum chemistry. Another sub-field of ML is deep learning (DL). By building a neural network that can simulate the human brain for analysis and learning, the mechanism of the human brain to interpret data can be imitated. Big data and associated data mining algorithm methods can provide researchers with the ability to discover new compounds that may be new drugs, discover or reuse drugs that may be more effective when used alone or in combination, and improve the field of personalized medicine based on genetic markers. With the increase in the amount of data and the continuous growth of computer performance, DL has gradually become a very important part of the AI sub-field, especially the flexibility of the neural network architecture it demonstrates, such as convolutional neural network (CNN), recurrent neural network (RNN) and fully connected feedforward network. It is believed that by combining with AI, the success rate of clinical trials can be continuously enhanced and a faster, lower cost and more effective drug development process can be created.

The role of AI in molecular target identification

In drug development, AI has changed the method of treating disease pathways or target identification by integrating genomic information, biochemical properties, and target tractability. A study showed that a computational prediction application using "Open Target", a data platform related to genetic diseases, can be used to predict the rationality of treatment targets. According to reports, the neural network classifier has a prediction accuracy of >71% for animal models showing disease-related phenotypes, providing the most effective prediction ability. There are a lot of examples in the application of AI in the drug development process. For instance, the AI platform IBM Watson for Drug Discovery has identified five new RNA binding proteins (RBPs) related to the pathogenesis of amyotrophic lateral sclerosis (ALS), a neurodegenerative disease.