Aro: a machine learning approach to identifying single molecules and estimating classification error in fluorescence...

Area of Science:

• Molecular Biology
• Biophysics
• Microscopy

Background:

• Advanced techniques enable single-molecule visualization, crucial for understanding biological dynamics.
• Fluorescence microscopy images often suffer from noise and background, hindering accurate molecule detection.

Purpose of the Study:

• To develop a computational pipeline for distinguishing true single-molecule signals from background noise in fluorescence microscopy.
• To enhance the accuracy and reliability of quantitative single-molecule experiments.

Main Methods:

• A computational pipeline was developed to analyze wide-field, epifluorescence microscope image stacks.
• The software employs a supervised random forest classifier to recognize and classify mRNA spots based on intensity features.
• It estimates the probability of true spots and quantifies classification error.

Main Results:

• The pipeline successfully distinguishes true single-molecule signals from background fluorescence and noise.
• Tested on challenging nematode embryo data, the software accurately identifies mRNA spots.
• Achieved >95% AUROC on artificial data and outperformed existing methods on real data.

Conclusions:

• The developed software effectively classifies single-molecule spots in challenging microscopy images.
• Provides statistically principled error estimation for improved data quality assessment.
• Offers confidence intervals for molecule counts, crucial for quantitative biological interpretations.