Research published today natural aging We described a machine learning algorithm to find senolytics and compared the algorithm’s findings to existing compounds.
The search for effective treatments
After covering the familiar area of senescent cells, this paper begins with a discussion of existing senolytic compounds, such as the well-known combination of dasatinib and quercetin. Most of these compounds were discovered through bioinformatic approaches investigating how these cells survive long after they should have died by apoptosis. [1].
Some anti-aging drugs have shown some efficacy in animal models, [2]there are relatively few of these drugs, and some have been found to be ineffective against the diseases they target, while others have serious side effects. It turns out that Lux lowers platelets and some immune cells. [3].
Even with these disappointing results in mind, researchers believe that senescent cells are viable targets and their removal with appropriate compounds may show clinical efficacy in treating the disease. We have discovered that the basic principles of generation have merit. However, finding these compounds is a problem.
Machine learning techniques are being used for drug discovery in other fields, including antibiotics [4]but this team points out that no progress has been made so far on using them to find senolytic substances. I tried to find them by training from scratch.
Working with very large datasets
The researchers employed an established machine learning graph model, trained with detailed information on 2,352 compounds, to demonstrate senolytic activity against human lung fibroblasts chemically aged with etoposide. Tested. 45 of these first 2,352 were selectively effective against senescent cells. The researchers then applied this trained model to 804,959 complete compounds.
The results returned by the model were highly variable. It turns out that some compounds are likely to be senolytic, while others are not. After filtering against , we selected 216 compounds selected by the algorithm and 50 compounds on hand as negative controls.
Preliminary analysis was favorable. Twenty-five of his first 216 were found to have senolytic properties in the real world by initial experiments. Although this is a relatively small percentage, it is clear that the algorithm effectively narrowed down a very large search space. Negative controls had no senolytic properties.
Comparison with the gold standard
The team then compared the efficacy of these candidates to ABT-737. ABT-737 has important senolytic properties, but its low bioavailability and side effects make it unsuitable for clinical use. At 10 micromolar doses, the researchers further narrowed the field down to his three specific compounds that were nearly as effective and specific as ABT-737. Importantly, none of these compounds decreased control cell viability, which ABT-737 is known to do.
Researchers have noted several attractive features of these compounds. They are drug-like compounds not currently in clinical use, dissimilar to those used in the training dataset, and their chemical properties make them suitable for oral administration. Further testing revealed that they did not overly harm healthy hepatocytes and were senolytic to doxorubicin-aged cells.
All three of these compounds have been found to bind to Bcl-2, a mechanism of action common to multiple senescent cell elimination. This is an important achievement in AI-based drug discovery. Despite the fact that they don’t look like other senolytic drugs, these algorithmically discovered compounds have been confirmed to work just as well.
Finally, when we tested one of these compounds in naturally aged mice, the results were positive. The treated mice did not appear to suffer any side effects, and the expression of the senescence markers SA-β-gal and p16 was significantly lower in the kidneys. Artificial intelligence has successfully discovered compounds that reduce key biomarkers of aging in animal models.
Conclusion
Of course, there are limitations to the models used in this study and the data used to train them. The initial training data consisted only of specific cell populations that were forced to age in specific ways. Senescent cells are highly heterogeneous, and other types of senescent cells may be vulnerable to completely different approaches that this trained model cannot discover.
However, this is an effective proof-of-concept and it certainly looks like AI drug discovery will be applied to senescent cell removal. A more robustly trained model on different types of senescent cells could provide even more useful information and uncover potential treatments that unassisted individuals might not have discovered on their own. Determining whether any of these drugs really work in humans is, as always, a matter of clinical trials.
literature
[1] Zhu, YI, Tchkonia, T., Pirtskhalava, T., Gower, AC, Ding, H., Giorgadze, N., … & Kirkland, JL (2015). to. senescent cells, 14(4), 644-658.
[2] Xu, M., Pirtskhalava, T., Farr, JN, Weigand, BM, Palmer, AK, Weivoda, MM, … & Kirkland, JL (2018). Senolytics improve physical function and extend lifespan in older adults . natural medicine, twenty four(8), 1246-1256.
[3] Rudin, CM, Hann, CL, Garon, EB, Ribeiro de Oliveira, M., Bonomi, PD, Camidge, DR, … & Gandhi, L. (2012). Monotherapy and biomarkers of Navitoclax (ABT-263) A phase II trial of is correlative in patients with recurrent small cell lung cancer. clinical cancer research, 18(11), 3163-3169.
[4] Stokes, JM, Yang, K., Swanson, K., Jin, W., Cubillos-Ruiz, A., Donghia, NM, … & Collins, JJ (2020). A deep learning approach to antibiotic discovery. cell, 180(4), 688-702.
