A robotic arm will take a single molecular fragment and, guided by machine learning, grow it into a drug candidate without human intervention. This matters because lead optimisation—the step where a weak starting molecule is chemically modified into a potent drug—is the most expensive and time-consuming phase of preclinical drug discovery. It relies on manual, repetitive steps that make it economically unfeasible to pursue treatments for many diseases, particularly neglected ones. The project automates this process using high-throughput experimentation, microscale reaction plates, and a feedback loop where machine learning algorithms analyse bioactivity data and decide which molecular modifications to try next. If the platform works, it could slash the cost and time required to identify drug candidates, making it viable to develop treatments for diseases that currently lack commercial incentive. The researchers will test the system against *Trypanosoma cruzi* prolyl oligopeptidase, a novel target for Chagas’ disease—a neglected tropical illness affecting roughly 6 million people in rural Latin America. Success would provide a proof-of-concept for fully autonomous drug discovery, with the potential to extend from simple binding optimisation to multi-objective drug design.
View original technical description
The most expensive and time consuming phase in preclinical drug discovery is lead optimization (LO) since it is comprised of several manual steps which are inefficient in terms of time and cost. These large expenses lead to many diseases being economically unfeasible to research hence this project aims to autonomise LO using High-Throughput Experimentation (HTE), labatory automation and machine learning through the introduction of Machine-Guided Molecular Growth (MG2). The MG2 workflow represents an automated feedback loop whereby starting materials with little biological activity are progressed into powerfully-binding drug-like molecules hence greatly accelerating bioactive molecule identification and significantly reducing costs. This will enable autonomous drug-candidate discovery for many diseases and help to eradicate suffering from potentially millions of patient worldwide. The workflow will take a single starting fragment and perform many reactions to produce bioactive molecules. Use of microscale plates will enable the efficient performance of hundreds of reactions simultaneously at a microgram scale, whose products will be screened against a biological target to measure bioactivity and most promising products identified. Machine learning algorithms will then determine the best molecular growth directions to further exploit biological activity and products of interest purified and dosed with further reagents ready for the next iteration of reactions and further functionalisation. The hardware set up will use a robotic arm to select starting materials and reagents from a compound library, where all solids are affixed to inert ChemBeads for ease of dispensing and liquids are pipetted, which then constitute one reaction within a 96-well reaction plate. Each reaction will then be purified using high-performance liquid chromatography-mass spectroscopy (LCMS), dispensed into a fluorescence bioassay and transferred to a fluorescence plate reader with the most promising reactions scaled-up and providing the starting points for further functionalisation. After the MG2 platform is built, a case study targeting Trypanosoma Cruzi Prolyl Oligopeptidase (TcPOP) enzyme, a novel target for Changas' disease, will be conducted. Changas' disease is a neglected tropical disease affecting rural areas of Latin America where poverty is widespread, currently affecting approximately 6 million people. The case-study will aim to improve binding affinity only, but with scope of multi-objective drug optimization, and serve as a proof-of-concept with goals to enable further drug optimisation in cellular assays.
Plain English summaries and category classifications on this site are generated by AI and may not perfectly reflect the original research.
Is something wrong? Let us know