MicrographCleaner

MicrographCleaner segments cryo-EM micrographs to identify and mask regions contaminated by carbon and other high-contrast artifacts, improving particle picking and preprocessing for single-particle cryo-EM analysis.

Key Features:

• Deep learning-based segmentation: Uses a U-net-like deep learning architecture tailored to segment cryo-EM micrographs into suitable and contaminated regions.
• Automated contamination discrimination: Distinguishes regions suitable for particle picking from those affected by carbon or other high-contrast contaminants to reduce false-positive picks.
• Training on curated dataset: The model was trained on a manually curated dataset comprising over five hundred micrographs.
• Benchmarking on independent micrographs: Performance and efficiency were evaluated through benchmarking on approximately one hundred independent micrographs.

Scientific Applications:

• Cryo-EM single-particle analysis preprocessing: Improves input micrograph quality by masking contaminated regions prior to particle picking.
• Particle picking accuracy: Reduces false-positive rates in automatic particle picking workflows by excluding high-contrast contaminants.
• Macromolecular reconstruction: Enhances downstream reconstructions of macromolecular structures by providing cleaner particle sets for structural biology studies.

Methodology:

Segmentation is performed by a U-net-like deep learning architecture trained on a manually curated dataset of over five hundred micrographs; the model discriminates regions contaminated by carbon or other high-contrast artifacts and was benchmarked on approximately one hundred independent micrographs to assess preprocessing efficacy and reduction of false-positive particle picks.