Making the Most of Nothing: One-Class Classification for Single-Molecule Transport Studies.

Area of Science:

• Single-molecule physics and chemistry
• Machine learning applications in scientific research

Background:

• Single-molecule experiments provide detailed molecular insights but generate significant noise.
• Stochastic nature of experiments leads to numerous blank traces, complicating data analysis.

Purpose of the Study:

• To evaluate the effectiveness of one-class (OC) classification for filtering blank traces in single-molecule experiments.
• To demonstrate the reliability of OC classification across diverse experimental conditions and datasets.

Main Methods:

• Utilized two distinct OC classification models.
• Trained models exclusively on blank traces from single-molecule transport experiments.
• Validated models on four diverse datasets from three laboratories, including cryogenic and room-temperature measurements.

Main Results:

• OC classification successfully filtered blank traces across all four tested datasets.
• Achieved high accuracy (96.9 ± 0.3%) and AUC (99.5 ± 0.3%) on a labeled cryogenic dataset.
• Demonstrated consistent performance regardless of temperature or laboratory origin.

Conclusions:

• OC classification is a powerful and reliable tool for enhancing single-molecule experiment data quality.
• This method significantly aids in the analysis of molecular properties by removing irrelevant signals.
• Facilitates a deeper understanding and manipulation of individual molecules.