A machine learning approach identifies cellular senescence on transcriptome data of human cells in vitro

Area of Science:

• Molecular Biology
• Bioinformatics
• Gerontology

Background:

• Cellular senescence is a recognized hallmark of aging.
• Detecting senescence cells (SnCs) is challenging due to molecular heterogeneity.
• Machine learning (ML) excels at identifying complex patterns in high-dimensional data.

Purpose of the Study:

• To evaluate the efficacy of ML algorithms in identifying SnCs.
• To compare the performance of different ML models using bulk RNA sequencing data.

Main Methods:

• Four ML algorithms (SVM, RF, DT, SIMCA) were tested.
• Bulk RNA sequencing data from 162 in vitro samples (fibroblasts, melanocytes, keratinocytes) were used.
• Senescence was induced by irradiation, bleomycin treatment, and replication.
• Validation included tenfold cross-validation, leave-one-out cross-validation, and independent dataset validation.

Main Results:

• All tested ML methods achieved over 80% accuracy in distinguishing SnCs.
• Support vector machines (SVM) demonstrated superior performance, reaching over 99% accuracy.
• Similar high accuracy was obtained using both algorithm-prioritized and expert-curated gene lists.

Conclusions:

• ML provides a powerful approach for identifying cellular senescence.
• SVM is highly effective for distinguishing SnCs from control cells.
• The study confirms the potential of ML as a proof-of-concept for senescence identification.