Discovering small-molecule senolytics with deep neural networks.

Area of Science:

• Biogerontology
• Computational Biology
• Drug Discovery

Background:

• Cellular senescence, characterized by the accumulation of senescent cells, is linked to aging, inflammation, and impaired cellular function.
• Senolytic drugs offer a therapeutic strategy to combat age-related diseases by selectively eliminating senescent cells.

Purpose of the Study:

• To identify novel senolytic compounds with potential therapeutic applications.
• To leverage deep learning for the prediction and discovery of senotherapeutics.

Main Methods:

• Screening of 2,352 compounds for senolytic activity against etoposide-induced senescence.
• Training graph neural networks to predict senolytic activity for over 800,000 molecules.
• Utilizing molecular docking and time-resolved fluorescence energy transfer assays to elucidate compound mechanisms.
• In vivo testing of a lead compound in aged mice.

Main Results:

• Identification of three structurally diverse, drug-like compounds with senolytic activity across various senescence models.
• These compounds demonstrated comparable or superior selectivity to the known senolytic ABT-737 and favorable medicinal chemistry properties.
• Mechanism of action partially involves the inhibition of Bcl-2, a key regulator of apoptosis.
• One compound, BRD-K56819078, significantly reduced senescent cell burden and associated gene expression in the kidneys of aged mice.

Conclusions:

• Deep learning approaches can effectively accelerate the discovery of novel senotherapeutics.
• The identified compounds represent promising candidates for treating age-related conditions by targeting cellular senescence.
• Further research into these compounds could lead to effective interventions for age-related diseases.