nanite: using machine learning to assess the quality of atomic force microscopy-enabled nano-indentation data

Area of Science:

• Biophysics
• Materials Science
• Computational Biology

Background:

• Atomic force microscopy (AFM) is crucial for mechanical characterization of biological samples.
• Force-distance (FD) curves in AFM experiments are often affected by artifacts like cell movement and probe adhesion.
• Manual sorting of artifact-laden FD curves is time-consuming and impractical for large datasets.

Purpose of the Study:

• To develop an automated method for analyzing and sorting AFM FD data.
• To implement supervised learning for distinguishing high-quality FD curves from artifacts.
• To create a reproducible and efficient data analysis pipeline for AFM.

Main Methods:

• Development of the Python package 'nanite' for FD data analysis.
• Implementation of supervised learning algorithms to classify FD curves based on extracted features.
• Quantitative analysis of Young's moduli for zebrafish spinal cord tissue.

Main Results:

• The 'nanite' package automates data import, baseline correction, contact point retrieval, and model fitting.
• Supervised learning achieves a mean squared error below 1.0 rating point for curve quality assessment.
• Classification accuracy for distinguishing good from poor FD curves exceeds 87%.

Conclusions:

• Automated quality-based sorting using supervised learning enhances AFM data analysis.
• The 'nanite' package provides a fully automated and reproducible pipeline for FD data analysis.
• Data quality can be integrated as a new dimension in quantitative AFM imaging.