AreTomoLive: Real-Time Cryo-Electron Tomography Alignment, Reconstruction, and Contrast Enhancement
1:17 Introduction to AreTomoLive: Developed at the Chan Zuckerberg Imaging Institute, AreTomoLive is a software package designed for real-time, high-throughput data processing in cryo-electron tomography (cryo-ET).
1:35 Transforming Cryo-ET: The goal is to make cryo-ET a routine method for probing cell biology, enabling visualization of molecular interactions within cells at subnanometer resolution.
2:15 Subtomogram Averaging (STA): This technique averages multiple copies of a molecule from tomograms to improve the signal-to-noise ratio and achieve higher resolution.
3:18 Throughput and Accuracy: AreTomoLive addresses the need for high throughput to match accelerated data collection rates and improve reconstruction accuracy by correcting for beam-induced motion and aberrations.
4:17 Contrast Enhancement: The software enhances contrast in cryo-ET data, which is crucial for downstream analysis like particle picking and segmentation.
4:28 AreTomoLive Workflow: Consists of two GPU-accelerated packages: AreTomo 3 for tomographic alignment and reconstruction, and Dooit for denoising.
6:09 Integration and Efficiency: AreTomo 3 integrates motion correction and tomographic alignment into a single pipeline, improving efficiency and consistency.
6:51 New Features in AreTomo 3: Includes multiple entry points for reprocessing, automated sample thickness measurement, local CTF estimation and correction, and quality metrics for data curation.
9:07 Importance of Sample Thickness: Accurate sample thickness measurement is crucial for optimal alignment and reconstruction quality.
9:42 CTF Estimation and Correction: AreTomo 3 estimates CTF by dividing tilt images into patches, rescaling power spectra, and refining parameters for local CTF correction.
10:45 Quality Metrics: AreTomo 3 provides quality metrics to assist in data curation, replacing time-consuming visual inspection.
12:46 Denoising with Dooit: Integrates denoising into the pipeline using a machine learning algorithm called noise-to-noise, which leverages a convolutional neural network with a U-net architecture.
16:06 Training Data Generation: AreTomo 3 generates paired noisy measurements for training the denoising model by splitting motion-corrected movies into even and odd frame sets.
17:59 Impact of Data Preprocessing: The way data is preprocessed, including data curation and CTF correction, significantly impacts denoising quality.
20:55 Leveraging Quality Metrics: AreTomo 3's quality metrics are used to select high-quality tomograms for training, ensuring robust denoising.
21:09 CTF Correction and Denoising: A more rigorous CTF correction, as implemented in AreTomo 3, improves downstream denoising by retaining more intermediate-resolution features.
25:37 Future Directions: Plans include leveraging denoised tomograms to improve tomographic alignment through subvolume polishing.
26:46 Team Effort: The development of AreTomoLive is a collaborative effort involving multiple individuals at CCI.
28:56 Model Generalization: The denoising model generalizes well to similar samples but may require retraining for different samples or data collection parameters.
38:44 Negative Control: Applying the denoising model to random noise has been tested and does not produce hallucinated features.