The new PiSCAT plugin, unsupervised anomaly identification (UAI), uses a deep neural network and classical machine learning to combine spatial-temporal features to track particles with low SNR in a background with noise structures similar to real iSCAT PSF (iPSF). The UAI deep neural networks are robust and fully automated algorithms that manipulate features of iPSFs and background speckles to the domain that can separate the footprint of the iPSF from background speckles. Since this method is unsupervised, it can be used for different applications with only fine-tuning of the hyperparameters. This method illustrates remarkable results in pushing the sensitivity limit in label-free detection of single proteins below 10 kDa.
Existing algorithms for detecting biomolecules using a label-free optical technique cannot detect single proteins with a molecular mass of fewer than 66 KDa with high precision [1, 2]. Although prior PiSCAT pipeline algorithms included spatial-temporal information, they operated in two steps with equal weights [2]. The modified version of this technique changed the localization part to reduce false positive detection by utilizing the capabilities of ML/DNN to mix spatial-temporal information with unequal weights in a simultaneous fashion. UAI demonstrates that an unsupervised machine learning approach for anomaly detection increases the mass sensitivity limit to below 10 kDa, i.e., by a factor of four compared to the state-of-the-art from other methods. Thus, this plugin enables the identification of tiny nanoparticles by optical methods. The details and findings of this pipeline are published in the journal Nature Methods with the title "Self-supervised machine learning pushes the sensitivity limit in label-free detection of single proteins below 10 kDa" [3].
Since this is a plugin for PiSCAT, first, you should install PiSCAT (see PiSCAT website); after that, this plugging can add to PiSCAT.
- To install PiSCAT using PyPi, enter the following command in the console (you need last version of piscat):
pip install piscat
- To add PiSCAT Plugins, enter the following command in the console in the same python environment you install piscat:
python -m piscat.Plugins UAI
Once the installation is done and the python environment is activated, enter the following command in the console:
python -m piscat.Tutorials
This plugin has two tutorials.
[1] Tutorial for anomaly detection (Hand crafted feature matrix)
[2] Tutorial for anomaly detection (DNN feature matrix)
This plugin has a separate license from the general PiSCAT. The license will be added very soon. Meanwhile, if you have any questions regarding the license, please contact with the following emails: