Towards AI-driven longevity research: An overview

Abstract

While in the past technology has mostly been utilized to store information about the structural configuration of proteins and molecules for research and medical purposes, Artificial Intelligence is nowadays able to learn from the existing data how to predict and model properties and interactions, revealing important knowledge about complex biological processes, such as aging. Modern technologies, moreover, can rely on a broader set of information, including those derived from the next-generation sequencing (e.g., proteomics, lipidomics, and other omics), to understand the interactions between human body and the external environment. This is especially relevant as external factors have been shown to have a key role in aging. As the field of computational systems biology keeps improving and new biomarkers of aging are being developed, artificial intelligence promises to become a major ally of aging research.

Keywords: artificial intelligence; biomarkers; deep aging clock; feature selection; longevity medicine; machine learning.

FIGURE 1

Summary of the hallmarks of aging. Primary hallmarks of aging, the primary causes of cellular damage, include genomic stability, epigenetic alterations, loss of proteostasis, and telomere attrition. Antagonistic hallmarks (this refers to factors that originated from body responses to the damage itself but end up exacerbating it) include mitochondrial dysfunction, deregulated sensing, and cellular senescence. Integrative hallmarks of aging (that result from the cumulative action of the previous two groups and are the main determiners of the functional decline) include stem cell exhaustion and altered intercellular communication. Each of these hallmarks has been the focus of intensive research to understand their involvement in the decline of biological functions. ML/AI technologies are used to deepen our understanding of the many components that are involved. This knowledge can help to improve not only our understanding of these mechanisms taken separately but also how the interplay between them unfolds.

FIGURE 2

AI-based automated CBMN test to quantify genomic damages in tissues. Thousands of images from flow cytometry data were used to train a DL-based image classifier. This allows an automated scoring of the evaluation of the number of micronuclei while reducing time overload and false positives.

FIGURE 3

Blood samples were utilized to identify epigenetic changes that allow distinguishing healthy controls from patients with coarctation of the aorta (CoA). Cytosine nucleotide (CpG) methylation changes obtained from Genome-wide DNA methylation analysis were selected as mean features. Only the probes with statistically significant methylation differences combined with a sufficiently high methylation fold change were kept as features and Multivariate approaches such as principal component analysis (PCA) and partial least squares-discriminant analysis (PLS-DA) were used to confirm that these features could accurately distinguish the CoA samples from the controls. 5 classifiers were built and trained using these features. Features with the highest predictive capabilities are potential novel biomarkers candidates for CoA.

FIGURE 4

To identify novel aging-related proteins, the full set of human proteins was extracted from the Swiss-Prot database. Already known aging related proteins listed on the GenAge database were used as instances of the aging-related class. UniProt, Gene Ontology and GeneFriends databases were used to extract 21,000 protein characteristics that were subsequently used as features to train 3 classifiers to identify proteins likely to be associated with aging.

FIGURE 5

To identify senescent cells, this study relied on the fact that senescent cells elicit a very specific morphology to generate a set of features using cell morphology images obtained by phase-contrast microscopy. The set of 50 × 50 pixels images were used to train a CNN as a classifier that could distinguish senescent cells from healthy cells with a greater accuracy than classical ML methods (SVM, RF, and LR).