Using Convolutional Neural Networks to Measure the Physiological Age of Caenorhabditis elegans

Area of Science:

• Biogerontology
• Computational Biology
• Machine Learning in Biology

Background:

• Caenorhabditis elegans (C. elegans) is a key model organism for aging research due to its short lifespan.
• Accurate physiological age measurement is crucial for antiaging drug and genetic screening.
• Current methods rely on subjective visual inspection, lacking precision for detailed aging studies.

Purpose of the Study:

• To develop quantitative methods for precise physiological age measurement in C. elegans.
• To utilize convolutional neural networks (CNNs) for daily age granularity, surpassing traditional period-based classifications.
• To improve the objectivity and accuracy of age assessment in C. elegans for aging research.

Main Methods:

• Utilized a dataset of 913 microscopic images of C. elegans across 14 days of adulthood.
• Applied five popular CNN models (ResNet50, InceptionV3, InceptionResNetV2, VGG16, MobileNet) for age prediction.
• Developed hybrid models combining CNNs with a 'curved_or_straight' attribute for linear and logistic regression analyses.

Main Results:

• Standard CNN models achieved an average testing Mean Absolute Error (MAE) of 1.58 days.
• The linear regression model integrating CNNs and the 'curved_or_straight' attribute achieved a test MAE of 0.94 days.
• The logistic regression model achieved 84.78% accuracy with a 1-day error tolerance.

Conclusions:

• CNN-based methods offer a quantitative and precise approach to measuring C. elegans physiological age.
• Hybrid models incorporating morphological attributes significantly enhance age prediction accuracy.
• These quantitative methods provide a robust tool for advancing antiaging research and drug screening.