One of the tasks in which machine learning and related techniques excel is finding patterns in very large data sets and extrapolating those patterns to predict as yet undiscovered members. The outcome of combinations of known small molecule drugs and drug candidates is one such data set. There is very little known in certainty about polypharmacology, as research and development groups operate under incentives that strongly discourage assessment of combination treatments. Where researchers have looked into combinations of small molecules in the context of slowing aging, they have found that the typical outcome is that any two compounds that individually alter metabolism to modestly slow aging produce no benefit or a mild harm when combined. This is a vast space of possibilities, little concrete knowledge, and maybe some useful outcomes hidden in the dross - and that is exactly the sort of challenge in which machine learning can be used accelerate the pace of discovery. That said, at the end of the day we are talking about effect sizes that are, at best, on a par with that of exercise. This isn't the path to radical life extension.
The genetic foundation of lifespan is becoming increasingly well-understood, but the optimal strategies for designing interventions to extend it remain unclear. Small molecule drugs, the mainstay of the pharmaceutical industry, act by modulating the activity of gene products - proteins, herein referred to as targets. Standard drug-discovery practice dictates that therapeutic compounds should be highly specific to a single target. However, closer inspection of FDA-approved drugs reveals that some of the most efficacious drugs bind multiple targets simultaneously and that, in some instances, more specific analogs are less efficacious. These findings suggest polypharmacology may improve efficacy for some complex indications.
The largest unbiased longevity screen of the Library of Pharmacologically Active Compounds (LOPAC), particularly FDA-approved drugs, identified a significant cluster of compounds that extend lifespan by modulating neuroendocrine and neurotransmitter systems. We observed that most inhibitors of G-protein coupled receptors (GPCRs) bind multiple structurally related targets, suggesting that polypharmacological binding increases their efficacy in extending lifespan. To test this notion, we used statistical and machine learning tools, specifically graph neural networks (GNNs), to identify geroprotector compounds that simultaneously bind multiple biogenic amine receptors and then evaluated their efficacy on the lifespan of Caenorhabditis elegans.
Over 70% of the selected compounds extended lifespan, with effect sizes in the top 5% compared to all geroprotectors recorded in the DrugAge database. Thus, our study reveals that rationally designing polypharmacological compounds enables the design of geroprotectors with exceptional efficacy.
Link: https://doi.org/10.1111/acel.70060
View the full article at FightAging
