LLAMA: Systematic Study of 4D Microscopic Datasets

Saptaparna Dasgupta, Bennett University

With recent microscopy improvements, cell-size data sets may be generated by capturing cell activity. The Lattice Light Sheet Microscope (LLSM) pictures cells at fast speed and high 3D resolution, collecting data over hours at 100 frames/seconds. The LLSM imaging gave fresh insights into the complicated immune cell surface behaviour, including the disclosure of novel ruffles. LLAMA (large-scale light microscopy analysis with machine learning) is a framework for the systematic study of the 4D microscopic datasets of the scale of terabytes. Recently, scientists from the University of Queensland employed a machine learning approach to partition semantic objects and created precise object-level statistics with a robust and customizable separation and algorithm tracking.

Lattice Light Sheet Microscope

The latest advances in microscopy include introducing an LLSM, a lattice light microscope that uses Bessel beam 2D optical lattice for resolution in an X, Y, and Z plane at a level with a high signal-to-noise ratio almost diffractive limit. 

The vertebrate cell’s surfaces could deform quickly to produce projections that interact with the surroundings. These surface projections comprise spike-like filopodia and wave-like ruffles on the macrophage surface when they do immunological monitoring. LLSM 3D records can track thousands of pulling cell surface surfaces over a long period of hours, routinely leading to terabyte-scale datasets that cannot be manually done, which require careful calibration or manual editing. 

Machine learning is a promising method that may be extrapolated through complex models with manually defined example data. These models may be used to execute solid tasks like classification and semantic segmentation at scale.

Tracking mechanism of LLSM

Semantic segmentation is intended to differentiate between structural classes and does not always separate individual items. An efficient LLSM data analytics system must thus add an object separation and tracking mechanism to the machine learning segmentation. Algorithms such as these are widely researched, but it should be guaranteed that the selected technique works well reliably on a scale, without the need for human editing. Macrophage ruffles are complicated and extremely dynamic and may be the worst situation of delineation and tracking of objects.

Deep learning models 

Deep learning models (in particular U-net models) offer a strong method for semantic segmentation. Sophisticated instance segmentation methods that mask R-CNN. It is based on a coevolutionary Neural Net Architecture, which has also been created for the direct identification of individual objects. Instead, Lefevre et al., 2021 have created a framework for quickly constructing and updating minimum data annotation models without the essential training information. This technique is particularly suited to instances where, due to experimental needs and markers, picture characteristics might change between datasets and various structural requirements can be determined.

Results from the study

Lefevre and his colleagues offered an analysis that illustrates the approach’s primary characteristics as well as its ability to generate novel biological results. Because the Trainable Weka method only requires a limited amount of training data, the background, cell body, ruffle, and filopodia segmentation classes were employed. On class-balanced data, the final model utilized the Weka random forest technique with default settings. Tenfold cross-validation was used to evaluate the model’s resilience on the training data, yielding a 0.73 per cent error rate. Cross-validated errors were consistently less than 1% after training on similar picture datasets. This model continues to have a very low error rate, indicating that it is quite stable.

The future of LLAMA

Machine learning algorithms are a promising approach that has shown to be capable of creating robust

semantic segmentations of microscope imagery. The team has created a unique visualization tool to aid in the

selection of training data and the evaluation of draught segmentations. The LLAMA image analysis system

was created to identify and understand features of cell membrane protrusions from large-scale data in a semi-

automated manner. The system may be deployed in a variety of ways, and cloud computing is supported.

The LLAMA pipeline is intended to be generally applicable, providing a method for identifying, monitoring,

and measuring structures that are not limited to the problem of macrophage surface projections.

Also read: Easy-Prime – A genome editing program

Source: Lefevre, J. G., Koh, Y. W. H., Wall, A. A., Condon, N. D., Stow, J. L., & Hamilton, N. A. (2021). LLAMA: A robust and scalable machine learning pipeline for analysis of large scale 4D microscopy data: analysis of cell ruffles and filopodia. BMC Bioinformatics, 22(1), 410. https://doi.org/10.1186/s12859-021-04324-z

• The Corrosion Prediction from the Corrosion Product Performance
• Nitrogen Resilience in Waterlogged Soybean plants
• Cell Senescence in Type II Diabetes: Therapeutic Potential
• Transgene-Free Canker-Resistant Citrus sinensis with Cas12/RNP
• AI Literacy in Early Childhood Education: Challenges and Opportunities

About the author: Saptaparna Dasgupta, currently a B. Tech 3^rd year student, pursuing Biotechnology, is a diligent student and determined in terms of her career goals. Being a budding biotechnologist, she is open to all research fields of her course and passionate about knowledge. She is focused and constantly tries to improve her writing skills, also a project enthusiast and is fond of gaining the hands-on experience in laboratories. She believes that all hard works and efforts pays off eventually and follows this as the motto of her life.