Deep Learning Helps Researchers Identify Bacteria in Seconds

Bacterial identification can generally take hours, often longer than that.

However, time is precious, especially when it comes to diagnosing an infection and choosing the right treatment.

Encouraged by this, researchers at the Korea Advanced Institute of Science and Technology (KAIST) trained a Deep Learning algorithm to identify the "fingerprint" spectrum of the molecular components of various bacteria.

Citing Eurekalert, the researchers were able to classify various bacteria on various media with up to 98 percent accuracy. Their research is published in the journal Biosensors and Bioelectronics.

Diseases caused by direct bacterial infection or by exposure to bacterial toxins can cause painful symptoms, even lead to death.

Therefore, rapid detection of bacteria is essential to prevent the intake of contaminated food and to diagnose infection from clinical samples, such as urine.

"By using Surface-Enhanced Raman Spectroscopy (SERS) analysis enhanced by the Deep Learning model, we presented a very simple, fast and effective route for classifying the signals of two common bacteria and their resident media without any separation procedure," said Professor Sungho Jo. from the School of Computing.

How to Lerka SERS
SERS sends light through a sample to see how the light spreads. The results revealed structural information about the sample – a spectral fingerprint – that allowed researchers to identify the molecule.

The surface-enhanced version places the sample cell on a precious metal nanostructure that helps amplify the sample signal.

However, it is difficult to obtain a consistent and clear spectrum of bacteria.

"In addition, the strong signal from the surrounding medium is also enhanced to overpower the target signal, and it requires a time-consuming and tedious bacterial separation step," said Professor Yeon Sik Jung from the Department of Materials Science and Engineering.

DualWKNet
The researchers applied Deep Learning which hierarchically can extract certain features from spectral information for data classification.

They specifically designed the model, which they named the Dual-branch Wide-Kernel Network (DualWKNet). The model was trained to efficiently study the correlation between spectral features.

Such capabilities, according to Professor Jo, are very important for analyzing one-dimensional spectral data.

DualWKNet enabled the research team to identify key peaks in each class that are barely visible in the individual spectrum, thereby increasing classification accuracy.

"Ultimately, with the use of DualWKNet instead of the bacterial and media separation step, our method dramatically reduces analysis time," say the researchers.

The researchers plan to use their platform to study more bacteria and media types. They will use that information to build a training data library of different types of bacteria on additional media to reduce collection and detection time of new samples.

"We hope to expand the use of this Deep Learnign-based SERS analysis platform to detect different types of bacteria in additional media that are important for food or clinical analysis, such as blood," said Professor Jo.

https://figshare.com/s/a06ee662d27a3caff3c7

https://figshare.com/s/db1d74d165c6733d88af

https://figshare.com/s/84de6802630c13eccc7a

https://public.flourish.studio/visualisation/8897781/ https://public.flourish.studio/visualisation/8897809/

https://safergreg4634.medium.com/

https://peatix.com/event/3189035/view?k=f3e8a36f8a6ce354b83bf3d29d0dd52517455056 https://peatix.com/event/3189044/view?k=d73524ac39a7ac6e54f78edbe9964f491eb4624b https://muckrack.com/1080/bio

https://figshare.com/s/dd5ecec4c11a963d8477

https://figshare.com/s/7dc281fd7802ff1ef3cd

https://public.flourish.studio/visualisation/8900276/ https://public.flourish.studio/visualisation/8900331/

https://crackedtwfilmpisan.tumblr.com/

https://crackecfullversionhdpisantw.tumblr.com/ https://www.blockdit.com/posts/6225f21d72d0b352e984f239 https://www.blockdit.com/posts/6225f2d1085531dbc1c254ab https://figshare.com/s/6abce416535cee2e225e

https://figshare.com/s/3b09422b5ddf124cca51