High-Throughput Nanorheology of Living Cells Powered by Supervised Machine Learning.

Area of Science:

• Biophysics
• Cellular Mechanobiology
• Materials Science

Background:

• Atomic force microscopy (AFM) is crucial for measuring cell nanomechanical properties.
• Current AFM data analysis, relying on model fitting, is slow and limits high-throughput mechanobiology.
• Existing methods face challenges with computational time and potential errors in data interpretation.

Purpose of the Study:

• To develop a supervised machine learning regressor for rapid analysis of AFM force-distance curves.
• To transform AFM data into nanorheological behavior efficiently.
• To overcome the limitations of low throughput in AFM-based mechanobiology studies.

Main Methods:

• A supervised machine learning regressor was developed and trained using theoretical AFM curves.
• Theoretical curves incorporated contact mechanics, power-law rheology, and bottom effect corrections.
• Functional data analysis was employed for regressor validation.

Main Results:

• The machine learning method reduced processing time for a large AFM dataset (2.62 × 10^5 curves) from hours to minutes.
• A 50-fold increase in throughput was achieved.
• The regressor accurately predicted cell modulus and fluidity coefficient with <4% relative error.

Conclusions:

• The developed machine learning regressor significantly enhances the throughput of AFM-based nanomechanical measurements.
• This approach offers a faster, more efficient alternative to traditional model fitting for AFM data analysis.
• The method provides reliable predictions of cell rheological properties, advancing mechanobiology research.