Researching and developing a new drug costs on average $1 billion while 90% of potential therapies fail somewhere between phase I trials and regulatory approval. The majority of experts in the sector feel that improving the efficiency of this process is imperative, which is one of the reasons why we’ve seen a significant increase in digitization in the pharmaceutical industry.
The relatively recent emergence of Artificial Intelligence (AI) has been met with enthusiasm, skepticism, and fear. Often hailed as a game-changer, AI holds untapped potential for improving traditional business practices and creating new opportunities for meeting and resolving major pain points across a variety of industries. This motivates its use in drug discovery and development since it can handle vast amounts of data with greater efficiency.
Artificial intelligence (AI) is a technology-based system that uses a variety of advanced tools and networks to emulate human intelligence, but contrary to popular belief, it does not threaten to replace human physical presence entirely. AI makes use of systems and software that can interpret and learn from data in order to make independent decisions and achieve specific goals.
It encompasses a variety of method domains, including knowledge representation, reasoning, solution search, and machine learning.
Machine learning or ML relies on algorithms to recognize patterns in data.
Deep learning, or DL, is a subfield of machine learning, and it relies on artificial neural networks (ANNs).
ANNs are sets of interconnected groups of nodes called perceptrons or artificial neurons that emulate human biological neurons and the transfer of electrical impulses between them.
ANNs can solve problems through either singly or multi-linked algorithms. Multilayer perceptron (MLP) networks, recurrent neural networks (RNNs), and convolutional neural networks (CNNs) are all examples of ANNs that use either supervised or unsupervised training techniques.
MLP networks are typically trained through supervised training methods that operate in a single direction. They can be used as universal pattern classifiers in pattern recognition, process identification, optimization aids, and controls.
RNNs, such as Boltzmann constants and Hopfield networks, are closed-loop networks that can memorize and store information.
CNNs are deep neural networks with one or more convolutional layers that rely on a set of dynamic systems with local connections. A convolution essentially means applying a filter over the input. The algorithm can take an input image, for example, and assign importance to different objects or aspects of that image which allows it to differentiate between them.
CNNs are mostly used in image classification, such as classifying handwritten letters or digits, but they can also be used for processing complex brain functions and biological system modeling. RBF networks, Kohonen networks, LVQ networks, counter-propagation networks, and ADALINE networks are some of the more complex types.
How Is AI Being Applied in Drug Discovery Efforts?
The cost of developing new pharmaceutical treatments has risen, as has the time it takes to bring them to the market. This is partly due to the increasing complexity of the illnesses being treated. Treatments for common illnesses have already been discovered, so the industry is now focused on rare and complex diseases involving complex chemistry with hundreds of proteins.
The process starts with selecting a target for the drug, which requires a thorough understanding of how the molecules will interact with each other. This is where artificial intelligence can play its part. Tech companies have developed software that can compare the biochemical and biophysical properties of millions of molecules and match them to the structures and properties of hundreds of thousands of proteins and find the molecules that are likely to bind to the target.
AI algorithms can be used to simulate over a billion interactions so they can narrow the list of compounds that may work. These tests also provide further information that helps refine the list. The process is repeated until researchers can reach a manageable list of candidate compounds, increasing time efficiency and reducing costs. More precisely, AI can reduce the time it takes to develop a new drug by up to 75% and reduce the costs by 80%.
AI’s ability to predict drug-target interactions can also be used to repurpose existing drugs. This reduces costs and time considerably because a repurposed drug can qualify directly for Phase II clinical trials. Machine learning is used to evaluate similarities between diseases, drugs, target molecules, and gene expression profiles.
Furthermore, these AI tests on drug-protein interaction can predict the likelihood of polypharmacology. Polypharmacology is a drug molecule’s tendency to interact with multiple receptors resulting in adverse effects. Thus, AI can facilitate the design of safer drugs by using Bayesian classifiers and SEA algorithms to go through vast databases and establish links between pharmacological drug profiles and potential targets.
When it comes to incorporating a new drug molecule into a suitable dosage form that offers the delivery characteristics scientists are looking for, AI can effectively replace the trial-and-error method through various computational tools that will evaluate stability issues, porosity, dissolution, and other factors.
Manufacturing a pharmaceutical product requires balancing a set of parameters. Conducting quality-control tests and maintaining consistency across batches typically involves manual interference, but in some cases, that might not be the most effective approach.
Once again, AI implementation can streamline the process. It can be used to analyze preliminary data from production batches and develop decision trees that will be further translated into rules that will guide operators in future production cycles. For instance, ANNs have been used to study and predict the dissolution profile of theophylline pellets with an accuracy rate of over 8%.
AI can also be used to regulate in-line production processes in order to reach the desired product quality standards. ANNs are used to track the freeze-drying process and estimate the temperature and desiccated-cake thickness at a later time point given a set of conditions, thereby assisting in the quality control of the final product.
The drug discovery process is well known to be time-consuming and costly. The digitization of data and technological breakthroughs have created an ideal setting for AI integration. It can be used at every stage of the process, from identifying potential targets and molecules to assisting with trial design and, eventually, continual post-market monitoring.
AI could revolutionize drug discovery and development, allowing the pharmaceutical industry to move away from traditional approaches and implement necessary changes that will increase quality and reduce costs, resulting in safer, more affordable treatments.