Aging is plastic: it can be slowed. But the search for drugs runs up against long experiments. Standard worm lifespan tests take weeks. Nearby in the archive sit papers on the links between genes and diseases, on peptides and antibodies, but our work stands apart: it combines machine vision with phenotypic screening. We speed up candidate selection using a simple trait, the worm's pose after stress.
Worms are grown to adulthood, placed 25 per well in a 96-well plate, and exposed to a drug for 24 hours. The plate is then clamped between copper blocks heated to 35°C for 11 hours. Survivors are photographed under a microscope. A Faster R-CNN neural network detects the worms, the SAM 2 model segments their silhouettes, and a classifier estimates the probability of survival from shape: a curled pose means alive, a straight one means dead. The average probability per well gives an estimate of survival.
The neural network distinguishes live from dead worms with an AUC accuracy of 0.88. The pose-based estimate correlates with manual counts of survivors (r=0.931). Of 2782 drugs, 130 passed the initial screen, and 31 were confirmed in a repeat test. 18 of them extended the worms' lifespan, and 14 had not previously been known as geroprotectors. The six strongest drugs increased lifespan by more than 30%. Three of them, icradipine, clemizole, and piperacillin, reduced the number of senescent cells in cultures of human endothelial cells.
This work differs from its neighbors in the archive: they study genes, peptides, or individual mechanisms, whereas here we have a holistic phenotypic screen with automated readout. For the first time, a worm's posture is used as a fast and scalable indicator of viability in drug screening. This opens the way to mass testing of compounds without expensive dyes or manual labor. The method can be transferred to other stress tests and disease models.
The method scales easily: the bottleneck is the size of the incubator and the number of copper blocks. With modest investment, thousands of compounds can be tested per week. The neural network can be fine-tuned to recognize other poses, such as paralysis, in order to search for drugs against neurodegenerative diseases. A transition to fully automated culturing and imaging systems is possible.
Pharmaceutical screening of geroprotectors, research on cellular aging, development of anti-aging therapies.
Test the selected drugs in mice. Study their mechanisms of action, especially the link between calcium channels and aging.
The key problem in gerontology is how to find substances that slow aging quickly and reliably. Our method cuts screening time from weeks to two days and does not require complex equipment. The question remains: how well do results in worms translate to mammals. We partially answer it by showing an effect on human cells.
The copper blocks clamping the plate eliminate the "edge effect," the difference in heating between the outer and central wells. Without them, survival in the outer wells would be lower simply because of uneven heat.
Not published on arXiv. Here are its page, the PDF and the formalization: you can check it yourself, not only read our account.