The journey of developing a new drug often takes years and needs hundreds of millions of dollars. A “shortcut” has now been found by a collaborated research led by City University of Hong Kong (CityU), which can potentially reduce the time and costs of developing new drugs by sorting out the high potential candidates out of a long list of chemical compounds. This breakthrough in neuropharmacology came following five years of collaborated research by CityU’s Department of Biomedical Engineering (BME), its Department of Biomedical Sciences (BMS), and Harvard Medical School. The research is published recently in the scientific journal Nature Communications, titled “High-throughput Brain Activity Mapping and Machine Learning as a Foundation for Systems Neuropharmacology”. Depression, psychosis, epilepsy and Alzheimer’s disease are some of the common brain disorders nowadays. But drugs designed to treat these serious ailments do not come easily due to the long developing timeline, and the particularly high failure rates* of many potential drugs before the right one is identified. The research, led by Dr Shi Peng, Associate Professor of BME at CityU, is to provide a platform to predict compounds that have the potential to be developed into new drugs to treat brain diseases. This platform can help drug developers to identify the compounds with a higher therapeutic and clinical translation potential, so as to prioritize the drug development pipeline and resources allocation. And more importantly, it can help speed up the new drug discovery process and save costs. “Even a 1% increase in the drug development success rate would make a huge difference for CNS disorder patients,” Dr Shi explained.
Innovative system for efficient whole-brain activity imaging
As in many other pharmacological research, this study used a small vertebrate animal, zebrafish, as a working model to conduct whole-brain activity mapping to show how and which part of the brain or central nervous system (CNS) react to the drugs. But Dr Shi said their innovative system helped to streamline the process, enabling large-scale experiments. “We have designed an integrative system that makes use of robotics, microfluidics and hydrodynamic force to trap and orient an awake zebrafish automatically in 20 seconds, instead of spending 20 minutes to prepare and manually position each single fish for a similar experiment. Therefore, we can carry out imaging for many zebrafish in one go to collect a large amount of data efficiently. Importantly, our platform is capable of immobilising the fish without anaesthesia, which may interfere with the brain activity and hence the evaluation of the chemical compounds,” he explained. By using this platform, the team firstly built a reference library of brain activity maps for 179 existing CNS drugs. They generated the maps from the brains of thousands of zebrafish larvae, each of which had been treated with a clinically used CNS drug respectively. The maps showed the corresponding brain regions that reacted to those drugs. Solely based on the intrinsic coherence among the maps of all the CNS drugs (without the names or any other information about the drugs), the team then used machine learning algorithms to classify these drugs into 10 physiological clusters. Some of the clusters were surprisingly discovered to be associated with the therapeutic categories, such as anti-epileptic, psychoanaleptic and anti-Parkinson, defined by the World Health Organization Collaborating Centre for Drug Statistics Methodology (WHOCC)**.
Machine Learning Help Predict Neuropharmacology
With the reference library in hand, in close collaboration with Dr Wang Xin, Assistant Professor of BMS Department at CityU and Dr Stephen Haggarty, Associate Professor at Harvard Medical School, the team went on to carry out information analysis and employed machine learning strategy to predict the therapeutic potential of 121 new compounds, by using the brain activity maps of these new compounds and those of the 179 clinically used drugs in the library. With a particular focus on anti-epileptics, the machine learning strategy predicted that 30 out of those 121 new compounds had anti-seizure properties. To validate the prediction, the research team randomly chose 14 from the 30 potential anti-seizure compounds to perform behavioural tests with an induced seizure animal model in zebrafish. “The result showed that 7 out of 14 compounds were able to reduce the seizures of the zebrafish without causing any sedative effects, implying a prediction accuracy of around 50%,” Dr Shi said. “With this high-speed in vivo drug screening system combined with machine learning, we can provide a shortcut to help identify new compounds with significantly higher therapeutic potentials for further development, hence speed up the drug development and reduce the failure rate in the process.” Another significant implication of the new screening paradigm is to make use of the physiology of zebrafish’s brain activity as an indicator of the therapeutic potentials of the compounds without the need of their biochemical information. “Traditionally, many drug development efforts were based on the study of the chemical structure or molecular target to identify potent compounds. But using our strategy, we actually found a large heterogeneity in chemical structures or molecular targets even within the drugs of the same classification of brain activity maps. Our new approach may help widen the pharmacology of certain neurological diseases,” said Dr Shi.
EurekAlert, 3 December 2018 ; http://www.eurekalert.org