Content overview: From glamorous stars to ordinary people in plain clothes, everyone will inevitably age, experience changes in description and degeneration of bodily functions. Because of this, people are also working hard to find the secret recipe for delaying aging. However, existing anti-aging drugs are always accompanied by some side effects. Recently, with the help of deep learning,
a research result published in "Nature Aging" screened out efficient and safe anti-aging drugs, which may bring us one step closer to "immortality".Keywords: Computational Modeling Machine Learning Aging
Author｜Xuecai
Editor｜Sanyang
This article was first published on the HyperAI Super Neural WeChat public platform~
Jonathan Swift once said, " Everyone wants to live a long life, but no one likes to grow old ." However, a study in "Nature Medicine" showed that at the three time points of 34, 60 and 78 years old, the gene expression of aging-related diseases in the human body will be up-regulated, leading to "cliff aging" in the human body. This also means that the aging of the human body may come earlier and faster than we think. How to stay young forever has once again become a focus topic.
In recent years, experiments have shown that the removal of senescent cells (Snc) in organisms through anti-aging drugs can improve the pathophysiological results caused by cell senescence and even prolong the lifespan of mice. However, these drugs have a series of side effects, including slowing wound healing, causing fibrosis of lung and perivascular cells, etc., and their efficacy and safety are difficult to balance.
To this end, Felix Wong et al. from the Massachusetts Institute of Technology (MIT) screened safe and efficient anti-aging ingredients from hundreds of thousands of compounds through deep learning and using a graph neural network, and tested them on mice Its efficacy and safety have been verified. The research results were published in "Nature Aging" in May 2023, titled "Discovering small-molecule senolytics with deep neural networks".
Paper address:
https://www.nature.com/articles/s43587-023-00415-z
Experiment overview
The researchers first screened drugs with anti-aging effects from some existing drugs, used them as training data for deep learning, and proposed indicators to measure their efficacy and safety. Then, based on the Chemprop model (a message propagation graph neural network model), they screened out highly effective and safe anti-aging drugs. After further screening, 3 compounds were obtained and compared with traditional anti-aging drugs for anti-aging properties and biological safety.
data set
The dataset for this study consists of two parts: 5,819 drugs from the Broad Institute's Center for Drug Repurposing, and 799,140 compounds from the Broad Institute's catalog.
experiment procedure
This experiment mainly includes 3 steps:
1. From the 2,352 drugs that have been approved by the US FDA and are undergoing clinical trials, the drugs with anti-aging effects are selected as the training set of the model;
2. Screen anti-aging drugs through the Chemprop model;
3. Compare the screened three representative compounds with the traditional anti-aging drug ABT-737 to verify their anti-aging properties and biological safety.
screening process
Anti-aging drugs need to meet the following three indicators:
1. The relative activity of normal cells after drug treatment > 0.7
2. Relative activity of senescent cells <0.5
3. The ratio of the activity of senescent cells to normal cells is less than 0.7
Based on these three criteria, the researchers first screened 45 drugs with anti-aging properties from the drugs approved by the FDA and in clinical trials, as the Chemprop model training set.
The Chemprop model exhibited extremely high drug selectivity, with an area under the precision-recall curve (PR curve) (AUC) of 0.24, which was significantly improved compared with the random model (0.019) and compared with the random forest model (0.15 ) also increased.
In view of the excellent performance of the Chemprop model, the researchers used Chemprop to screen the compounds in the data set. Among them, 284 of the drugs included in the Drug Repurposing Center of the Broad Institute had a predicted value higher than 0.1 . Of the compounds included in the Broad Institute, 2,537 compounds had predicted values (PS) higher than 0.4, and 3,838 drugs had very low predicted values, indicating no anti-aging properties.
Green: drugs that may have anti-aging properties in the Broad Drug Repurposing Center (PS>0.1);
Black: Compounds listed by the Broad Institute that may have anti-aging properties (PS>0.4);
Yellow: Compounds that have been verified to have anti-aging properties in later stages;
Purple: compounds predicted to have no antiaging properties;
Red: compounds with anti-aging properties in the training data;
Blue: compounds with no anti-aging properties in the training data.
Based on the chemical structure and pharmacokinetic properties, the research team further screened these compounds. First, pan-screened interfering compounds (PAINS) and adverse pharmacokinetic chemicals were removed. They then selected 216 compounds with Tanimoto similarity less than 0.5 for structural distinction from known antiaging drugs. At the same time, they also selected 50 drugs without anti-aging properties as negative controls. Finally, the researchers verified the anti-aging properties of these 266 compounds by chemical means.
Of the 216 high-scoring compounds, 25 exhibited antiaging properties in experiments. The positive prediction rate of Chemprop model was 11.6%, which was higher than 1.9% of manual screening. However, none of the 50 negative control compounds had anti-aging properties, indicating that the Chemprop model performed well in negative prediction.
Compare and verify
After obtaining the target compound, the researchers compared it with existing anti-aging drugs. Firstly, human lung fibroblasts (IMR-90) treated with etoposide (IMR-90) were used to construct the senescent cell model, and then the cells were treated with the screened BRD-K20733377, BRD-K56819078 and BRD-K44839765 respectively, and Compared with the conventional ABT-737 drug.
From the results, we can see that the compounds screened by the graph neural network have a good removal effect on senescent cells, and at the same time do not affect the growth of normal cells, and have strong selectivity. On the contrary, while ABT-737 cleared senescent cells, it also killed some normal cells, with strong side effects.
Gray: normal cells in the control group;
Blue: senescent cells obtained after etoposide treatment.
Subsequently, the researchers performed replicative senescence experiments with IMR-90 cells at the early and late passages and obtained similar results. Further, they performed hemolysis experiments to test the biological toxicity of these drugs. The results showed that even when the dose of the drug reached 10 times the normal dose (100 μM), hemoglobin released by red blood cell death was hardly detected in the blood, proving its biological safety.
Based on the above results, the researchers conducted in vivo experiments on C57BL/6J mice with BRD-K56819078, which has the most cell selectivity. After 14 days of drug injection, the kidney cells of the mice were collected to observe the content of senescence-associated β-galactosidase (SA-β-gal) and the expression of related mRNA.
a: SA-β-gal content;
b: The expression level of senescence-related mRNA
The results showed that both SA-β-gal content and mRNA expression were down-regulated, indicating that BRD-K56819078 effectively cleared senescent cells in mice. After layers of screening, the Chemprop model finally obtained an efficient and safe anti-aging drug.
Chemprop Model: A Good Helper for Drug Development
The Chemprop model is a deep learning model based on a graph neural network (GNN). It has 5 layers, 1,600 hidden dimensions, and is more complex than ordinary GNN models.
An eigenvector is generated in Chemprop for each atom and chemical bond based on the following characteristics:
1. Atomic characteristics such as number of atoms, number of bonds per atom, formal charge, chirality, number of bonds with hydrogen atoms, hybridization, aromaticity, and atomic mass;
2. Bond type (single bond, double bond, triple bond or aromatic ring, etc.), conjugation, whether it forms a ring, and three-dimensional characteristics of chemical bonds.
The Chemprop model utilizes a message propagation convolutional neural network to characterize compounds. By summing up the messages of adjacent bonds, then comparing it to the total bond sum, and finally processing it with a single neural network layer with a non-linear activation function, we can get the message of a chemical bond. After a fixed number of message passes, the messages for the entire molecule are summed to obtain a message value representing this molecule. After feeding this value into the feed-forward neural network, the Chemprop model will output a predicted value that correlates with the activity of the compound.
At present, the Chemprop model has been widely used to predict the drug activity of compounds and to screen and develop new drugs.
In 2020, MIT used Chemprop to screen out 8 antibacterial drugs with different structures from existing antibiotics from more than 107 million molecules, and found Halicin, a drug molecule that exhibited broad-spectrum antibacterial activity in mice. In 2022, the research team of Capital Medical University used Chemprop to screen out a possible inhibitor of cathepsin L, providing a new target for killing the new coronavirus.
Reference link:
[1] https://doi.org/10.26434/chemrxiv-2023-3zcfl
[2] https://www.cell.com/cell/fulltext/S0092-8674(20)30102-1
[3] https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9110316/