Bacteria were identified in seconds with a "Fingerprint" machine learning technique

Bacteria were identified in seconds with a "Fingerprint" machine learning technique ...

When diagnosing infections and selecting appropriate medications, bacterial identification may take hours and often longer, depending on whether or not they are suitable. KAIST proposes a more precise and quicker process. By developing a deep learning technique to identify the molecular components of several bacteria, researchers may classify various bacteria in different media with a certainty of up to 98%.

In Biosensors and Bioelectronics, their results were available online on January 18th, before they will be published in the journal April issue.

Bacteria-induced illnesses, caused by direct bacterial illness or by exposure to bacterial toxins, can induce painful symptoms and lead to death, thus a rapid diagnosis of bacteria is crucial to prevent the intake of harmful foods and to detect infections from clinical samples, such as urine. According to Professor Sungho Jo, the school of computing has developed a markedly simple, quick, and effective way to classify the signals of two common bacteria and their resident media without any separation procedures.

The Raman spectroscopy provides information about the sample''s structure, which allows researchers to identify its molecules. The surface-enhanced version focuses on noble metal nanostructures that help amplify samples signals.

Due to a number of overlapping peak sources, such as proteins in cell walls, it is difficult to obtain consistent and clear bacteria spectra. Moreover, strong signals of surrounding media are enhanced to overwhelm target signals, requiring lengthy and tedious bacterial separation steps, according to Professor Yeon Sik Jung from the Department of Materials Science and Engineering.

According to Professor Jo., researchers used deep learning to systematically extract certain features of the spectral information to classify data. Their use, which is called the Dual-branch wide-kernel network (DualWKNet), was developed to help identify the correlation between spectral features. Such an ability is critical for analyzing one-dimensional spectral data.

Despite having interfering signals or noise from the media, which make the general shapes of different bacterial spectra and their residing media signals look similar, Professor Jo said the high classification accuracies of bacterial types and their media were discovered, owing to the possibility that DualWKNet was able to detect key peaks in each class that were almost indiscernible in individual spectra, reducing the analysis time.

In addition to media, researchers intend to use their platform to study more bacteria and media types, utilizing information to develop a training data library of bacterial types to reduce the collection and detection times for new samples.

Professor Jo said that with the collaboration between SERS and deep learning, we wanted to extend the use of our deep learning-based SERS analysis platform to detect numerous kinds of bacteria in additional media that are important for food or clinical analysis.

You may also like: