What it's about: Research links cellular senescence to diseases like cancer, type 2 diabetes, osteoarthritis, and viral infections. Although drugs to clear senescent cells have gradually become a research and development hotspot. However, due to the lack of well-characterized molecular targets, few anti-aging compounds (Senolytics) have been discovered. Recently, the international journal "Nature Communications" published a research result in which researchers discovered three new types of Senolytics.
Keywords: Senolytics machine learning XGBoost
This article was first published on the HyperAI super neural WeChat public platform~
Since ancient times, people have been pursuing immortality. Surprisingly, in recent years, topics such as anti-aging and longevity are moving from the mysterious and illusory health care products industry to the medical and health industry recognized by the public. In general perception, aging is the process of slow weakening of body functions, and the process is irreversible, so human beings can only let nature take its course and resign themselves to fate. However, what many people don’t know is that as early as 2018, the World Health Organization (WHO) declared aging as a treatable disease in the International Code of Diseases.
In the broad definition of aging, cellular aging is one of the hot research directions of scientists recently. Cellular senescence is a phenomenon characterized by the cessation of cell division. Normally, the human immune system can effectively clear senescent cells, but with age, this clearing function will gradually weaken. In addition to causing vision deterioration and limited mobility, it is also very easy to cause cancer, Alzheimer's disease and other diseases.
In 2015, Dr. James L. Kirkland of the Mayo Clinic and others discovered the first anti-aging drug (Senolytics) that can clear senescent cells. Senolytics refers to a small molecule compound that selectively induces the death of senescent cells. Its name comes from Senescence (aging) and Lytic (destruction). In the latest study, the University of Edinburgh and the University of Cantabria used machine learning to discover three Senolytics - Ginkgetin, Periplocin and Oleandrin, and verified their anti-aging effects in human cell lines. The research has been published in the journal Nature Communications under the title "Discovery of Senolytics using machine learning".
Figure 1: The research results have been published in Nature Communications
Paper address:
Discovery of senolytics using machine learning | Nature Communications
experiment procedure
data set
This experimental dataset comes from multiple sources, including academic publications and commercial patents. First, the researchers mined 58 known Senolytics, and then mined a variety of non-Senolytics from two existing chemical libraries, LOPAC-1280 and Prestwick FDA-approved-1280. The dataset combines the two and contains a total of 2,523 compounds, of which Senolytics accounted for 2.3%.
Figure 2: Compounds used to train machine learning models
a: The training data comes from multiple channels.
b: 58 Senolytics sources used for training, including the number of compounds and cell lines for each source.
model training
The researchers used the aforementioned dataset to train a model to identify compounds with senolytic (positive) signatures. First, the researchers performed feature selection on the dataset, during which they calculated the average Gini index reduction for each feature using a random forest (RF) model, selecting the 165 most important features, thereby reducing the number of features and reducing model complexity.
- The Gini index measures how mixed the samples in a node are, and the lower the value, the purer the samples in the node.
Second, the researchers developed multiple senolytics binary classification models (identifying senolytics or non-senolytics) using the 165 most important features as well as various sample data from the full dataset. In order to compare the models and take full advantage of the limited number of Senolytics samples, the researchers performed 5-fold cross-validation on the dataset, scoring the models using 3 performance metrics: precision, recall, and F1 score.
Initially, researchers focused on support vector machines (SVM) and RF models, but experiments found that neither performed well. They also evaluated other models of varying complexity, including Logistic regressors, Naïve Bayes classifier, and Data Augmentation for Imbalanced Classification (SMOTE), but the results showed that the performance of these models was not as good as that of SVM and RF models.
Therefore, using RF performance as a benchmark, the researchers developed the XGBoost model to improve predictive power by iteratively training a decision tree model. As shown in Fig. 3b, the XGBoost model improves in precision, recall and F1 score, and performs best among all considered models.
Figure 3: Training a machine learning model
a: Model training, compound screening and result verification process, using multiple performance indicators to select suitable models.
b: Performance of 3 machine learning models, bar graph shows average performance metrics computed in 5-fold cross-validation, error bars represent one standard deviation.
The address of this experimental data set will be synchronized to HyperAI's official website later:
Code and data for "Discovery of senolytics using machine learning" | Zenodo
Experimental results
First, the researchers screened more than 4,340 compounds for 21 compounds that might have antiaging activity. They then tested these 21 compounds. As shown in Figure 4, three of them have senescent cell-clearing effects: Periplocin and Oleandrin (two cardiac glycosides that have not been identified to clear senescent cells) and Ginkgetin (a natural non-toxic biflavonoid compound).
Figure 4: Periplocin, Oleandrin and Ginkgetin have the effect of clearing senescent cells
c: Experimental verification. Three out of 21 compounds showed anti-aging activity: Ginkgetin, Oleandrin, and Periplocin; heatmap shows mean of n = 3 replicates. Ouabain in the figure is a known Senolytics.
d: Dose-response curves of 3 newly discovered anti-aging compounds. SI is the anti-aging index.
In addition, during the experiment described above, the researchers also found that the newly discovered Oleandrin had stronger anti-aging properties than Ouabain, especially at low concentrations. Therefore, the researchers compared the antiaging activities of Periplocin, Oleandrin and Ouabain at a low concentration of 10 nM.
Figure 5: Comparison of anti-aging properties of Periplocin, Oleandrin and Periplocin at low concentrations
a: Tissue culture dishes of IMR90 ER:RAS (senescent cells) and IMR90 ER:STOP (control group) cultured with 100 nM 4OHT are shown. Treatment with 10 nM Oleandrin, Ouabain, and Periplocin and DMSO (control) was performed for the next 72 hours.
b: Evaluation of cell viability by quantitative analysis.
As shown in Figure 5b, low concentrations of Ouabain and Periplocin did not show significant cytotoxicity in IMR90-ER:STOP and IMR90-ER:RAS, but after treatment with Oleandrin, the survival rate of senescent cells in IMR90-ER:RAS was significantly decreased, indicating that Oleandrin has strong anti-aging activity at lower drug concentrations . Based on the above experimental results, machine learning can successfully find anti-aging compounds, and also find Oleandrin with stronger anti-aging properties than existing anti-aging compounds.
AI-Driven Drug Discovery
AI plays an important role in all stages of new drug development. Currently, the research focus is on the drug discovery and preclinical development stages. This study demonstrates the potential of AI in drug discovery, especially when addressing diseases with complex biological structures or few known molecular targets. Author Diego Oyarzún states: "AI is very effective in helping us discover new drug candidates, especially in the early stages of drug discovery."
Vanessa Smer-Barreto, the first author of the study, emphasizes the importance of close collaboration between data scientists, chemists and biologists. "This work arose through close collaboration between data scientists, chemists, and biologists. We took advantage of this interdisciplinary collaboration to build a robust model and save screening costs by using only published data for model training," she said. This collaborative model offers new opportunities to accelerate AI applications and is expected to drive innovation and advancement in drug discovery and development .
At present, although AI has made breakthroughs in the development of new drugs, it still faces some challenges, such as data quality and reliability, algorithm interpretability, and model generalization capabilities. With the continuous advancement of technology and the increase of data resources, the application prospect of AI in drug development is still very broad. By strengthening data sharing and interdisciplinary cooperation, the advantages of AI can be better utilized to accelerate the discovery and development of new drugs and bring benefits to human health.
Reference article:
[1]http://zixun.69jk.cn/shwx/79532.html
[2]https://en.wikipedia.org/wiki/Cellular_senescence#Characteristics_of_senescent_cells
[3]https://newatlas.com/medical/machine-learning-algorithm-identifies-natural-anti-aging-chemicals/
[4]https://www.sohu.com/a/673349496_121124375
[5]https://www.ed.ac.uk/institute-genetics-cancer/news-and-events/news-2023/ai-algorithms-find-drugs-that-could-combat-ageing
[6]http://www.stcn.com/article/detail/904319.html
This article was first published on the HyperAI super neural WeChat public platform~