Transforming Drug Discovery Using AI and Automation

Read Time:

The COVID-19 pandemic has unveiled a pressing issue – the need to develop effective drugs rapidly. But developing a new drug is easier said than done. Drug discovery begins with a hypothesis that the inhibition or activation of a target molecule or pathway results in a therapeutic effect. After target identification and validation, comes hit-to-lead and lead optimization steps. This involves identifying hit molecules with an affinity to the target and selecting the “best” molecule to take forward.

Lead optimization is an elaborate process that can be costly and time-intensive; despite the resources devoted to generating lead compounds, there is always a moment of uncertainty that a drug will progress to the next phase of development. Due to this complex and uncertain pathway, there has been considerable interest in deciphering ways to improve the success rate in drug discovery, and one such means is automation. 

Automation encompasses innovations that can transform the drug development process. Throughout the drug discovery value chain, automation has the potential to increase laboratory efficiency, lower overall attrition, and decrease costs. New technologies such as microfluidics, robotics, and the use of artificial intelligence, combined with automated data analysis, can expedite the drug development and approval process, helping make therapies available to patients more quickly.

High‐throughput screening – Delivering key starting points in drug discovery

High-throughput screening (HTS) involves using automated equipment to rapidly evaluate the activity of large numbers of compounds against specific biological target. “The main advantage of using high-throughput screening in lead discovery is the ability to test thousands to hundreds of thousands of agents (small molecules or functional genomics tools) in a rapid and reproducible manner,” says Charles Karan, scientific director, Columbia Genome Center High Throughput Screening facility. HTS can be viewed as a fast scan of biological processes by which candidates with inadequate or no effect can be rapidly excluded from the drug discovery pipeline.

HTS labs employ a variety of assay formats, and the automation of assays plays a central role in the process. “The goal of these assays is to take the large collection of agents and whittle them down to the small number that shows promising results in the assay,” explains Karan. Automation is achieved by employing liquid handling devices, robotics, plate readers as detectors and dedicated software for instrumentation control and data processing. 

Automation allows for testing greater numbers of hypotheses

Apart from reduced costs and decreased timelines, automation improves data accuracy, precision, reproducibility and traceability; and these factors allow researchers to exploit high‐quality data in hypothesis‐driven research. “Due to the precision of the automation, all of the plates in the assay will be run under very similar conditions and by using plate-based controls we can assure assay uniformity throughout the screen,” explains Karan. Furthermore, automation allows researchers to test greater numbers of hypotheses and can enable complex workflows and screening scenarios that may be difficult or impossible to achieve manually.

Robotics – Improving accuracy and reproducibility

Robotics improve the overall efficiency of a process by creating efficient means of completing pre-set tasks. Robots are relentless systems capable of parallel processing; they can manage multiple sequential steps in any workflow simultaneously without halting or "taking breaks”. Another factor that makes robots the heroes of automation is precision – robotic systems can remarkably increase the accuracy and reproducibility of the process and quality of data capture, which is hard for a researcher to achieve manually. 

According to Karan, “Data accuracy comes from uniformity of assay. The obvious advantage to using automation and automated liquid handling is that crucial steps of dispensing very small volumes are handled by highly calibrated liquid handling instruments.  There are two things that we do to ensure this, first, we validate the individual instruments liquid-handling accuracy and secondly we monitor the results of each assay plate (which includes controls), to ensure that the signal for the assay does not change overtime. This includes adding a pass/fail metric for each plate to confirm data quality.”

Microfluidics-based miniaturized discovery platform

Microfluidics has generated significant interest in the drug discovery and development domain. Microfluidics offers the distinct advantage of system miniaturization. By modulating the movement of minute quantities of fluids, microfluidics helps miniaturize assays and increase experimental throughput. The past few years have witnessed a rapid rise in the use of microfluidic technologies, such as 3D cell culture systems, organ-on-a-chip and lab-on-a-chip technologies, and droplet techniques. 

The volume of the microfluidic device is minimal, and many functions can be integrated on a single chip. The internal dimensions of the chip range from micrometers to millimeters; this allows the handling of samples and reagents even in the picoliter range. Microfluidic chips, coupled with multichannel and array designs, allow a high-throughput process to be achieved, increasing the speed at which you can screen. Apart from rapid screening and analysis, microfluidics lowers reagent consumption and costs by virtue of its miniaturized devices. 

A promising technology utilizing microfluidic technology is organ-on-a-chip. Organ-on-a-chip refers to a physiological organ biomimetic system built on a microfluidic chip. Because of their ability to closely mimic the dynamic interactions of in vivo microenvironments, organ-on-a-chip and body-on-a-chip systems hold great promise for the development of high-throughput assays that could be valuable in drug screening and toxicity studies.

Integrated Biologics LIMS Designed To Accelerate Drug Discovery Efforts With the increasing number of large molecule therapeutics in the pipeline, biopharmaceutical companies needed to change the way they discover and develop new treatments. Large molecule discovery and development is an intricate high-throughput process – the amount of data produced and analyzed has increased rapidly, necessitating much improved laboratory information management systems (LIMS). Download this case study to discover a comprehensive, highly integrated, biologics LIMS solution. View Case Study Sponsored Content

Artificial intelligence in drug discovery

Modern biology is increasingly rich in data, such as the vast amount of genetic data that has led to the generations of thousands of genomic databases. However, these large datasets require appropriate analytical methods to yield statistically valid models that can make predictions. Artificial intelligence (AI) can be exploited to capture these large datasets and utilize them for early target identification and validation. 

AI refers to the ability of a machine (such as a computer) to perform tasks in response to a range of environments. Machine learning (ML), a subset of artificial intelligence, uses algorithms that can learn and improve without reprogramming. To predict outcomes related to drug discovery, the machine requires algorithms to process existing data and identify patterns of properties. ML can be used in many stages of drug discovery. Validation of biological targets, discovery of drug candidate molecules, identification of biomarkers predictive of disease are some examples of areas where ML can be utilized. 

As per Dean Ho, PhD, Provost’s chair professor at the National University of Singapore: “Drug discovery involves the identification of novel drug structures/compounds. This can be markedly accelerated using AI, and the ability for that novel compound to engage its target can also be improved and accelerated using AI.”

By using AI, a virtual compound library of several billion molecules can be screened, and preclinical candidates can be identified in a shorter time frame than conventional methods. John Mitchell, PhD, theoretical chemist at the University of St. Andrews, says, “AI can help to make the drug discovery process more effective by doing lots of computations very fast, though not necessarily very accurately.” Mitchell enumerates that AI can assist drug discovery by:

• Linking genes to diseases
• Identifying proteins as therapeutic targets against these diseases
• Exhaustively evaluating the possibility of repurposing each known drug against every possible target for each known disease
• Quickly screening databases of millions of possible or existing drug-like molecules as possible medicines, eliminating those with poor ADMET (absorption, distribution, metabolism, excretion, and toxicity) and solubility (by doing fast chemoinformatics), identifying those which may be active (possibly by fast docking or pharamacophore searches)
• Prediction of possible side effects
• Linking the efficacy of therapeutics to human genetic variation both at the group and the individual level

Enhancing the path from drug concept to patients

Describing the role of AI in drug discovery, Ho says, “Firstly, it is important to note that there are multiple segments to bringing a drug to patients. Broadly speaking, they include drug discovery, drug development, and drug administration. These segments are often viewed as the same, but in fact, they are very different domains that can all be enhanced via AI, and must all be addressed properly and then seamlessly integrated in order to truly optimize how patients are treated.”

Dean elaborates further, “With regards to drug discovery, there is often a perception that a well-designed molecule will ultimately lead to a successful outcome and will make it to market. This is far from a guaranteed outcome since even promising drugs delivered at sub-optimal doses or in combination with the wrong companion therapies (combination therapy) can lead to – at minimum – a suboptimal outcome, or commonly, poor outcomes and ultimately trial failure. As such, even good drugs delivered incorrectly will not work out. This is where drug development comes in. Drug development includes the processes of determining how to best co-deliver different drugs together as combinations, how to match patients to the right clinical trials to improve their chances of responding to treatment, and how to design clinical trials to best determine if the drug in fact improves outcomes. Emerging AI platforms (e.g. IDentif.AI, CURATE.AI, QPOP, etc.) are demonstrating that drug combination design can be optimized from extraordinarily large drug and dose spaces.”

Automation in drug discovery – The hurdles 

Biological systems are complex sources of information that are now being systematically measured and mined at unprecedented levels using a plethora of innovative technologies. While there are many observed benefits of implementing automation in drug discovery, there are significant impediments to its utilization. AI is a data-mining method – despite its specific application, systematic and comprehensive high-dimensional data needs to be generated and stored in a centralized manner. The quality of available data is also important; before performing specific AI tasks, refining the raw inputs for high-quality data is crucial. The need for investment to access AI technology is yet another challenge.

“AI-based drug discovery is promising, but there is a long way to go to ensure that these newly discovered molecules are developed properly. The different segments have to be united. In terms of downstream challenges, it is critical to unite stakeholders early on,” explains Ho.

Dr Ho also explains that “the field of AI and AI-based tech development will have to evolve to align with the changing needs in therapeutics. At some point, we’ll be able to move beyond solely relying on pre-existing data and algorithm training and prediction making.”

Adding a different perspective to the challenges of AI in drug discovery, Mitchell says, “There are plenty of (often smaller) companies and even academics doing this stuff already. The technical challenges are already being met. The most obvious risk is excessive hype and expectation. Technologies such as combinatorial chemistry and chemoinformatics have been initially over-hyped and subsequently underappreciated. Best practice use of AI shifts the odds favorably, but isn't a guarantee of immediate success. Drug discovery is an inherently risky and expensive business.”

The application of lab automation, robotics, and AI in drug discovery has taken tremendous strides in the past few years. Though the road to success is challenging, automation in drug discovery can help make quicker decisions and enable lifesaving drugs to reach the right person at the right time.