Bacterial detection can take hours, if not days, to complete—wasting valuable time for diagnosing diseases and choosing treatments. According to KAIST scientists, there may be a faster and more precise method. The researchers were able to categorize bacteria in diverse environments with 98% accuracy by instructing a deep learning system to recognize the “fingerprint” spectra of the chemical components of individual bacteria.
Schematics of the general process of Raman data collection and analysis where a single spectrum is attained from a single cell and classified via deep learning. Image Credit: Korea Advanced Institute of Science and Technology.
Their findings were published online in the journal Biosensors and Bioelectronics on January 18th, 2022, ahead of publication in the journal’s April edition.
Bacteria-induced ailments, such as those due to the direct bacterial infection or exposure to bacterial toxins, can cause severe symptoms and even fatality. Hence, early diagnosis of bacteria is essential for avoiding contaminated food consumption and diagnosing infections from clinical samples like urine.
By using surface-enhanced Raman spectroscopy (SERS) analysis boosted with a newly proposed deep learning model, we demonstrated a markedly simple, fast, and effective route to classify the signals of two common bacteria and their resident media without any separation procedures.”
Sungho Jo, Professor, School of Computing, Korea Advanced Institute of Science and Technology
Raman spectroscopy examines how light scatters through a substance. The spectral fingerprint provides structural information about the sample, helping researchers to discover its constituents. Sample cells are placed on noble metal nanostructures in the surface-enhanced form, which helps enhance the sample’s signals.
Due to various redundant peak sources, such as proteins in cell walls, it is difficult to get consistent and accurate spectra of bacteria.
Moreover, strong signals of surrounding media are also enhanced to overwhelm target signals, requiring time-consuming and tedious bacterial separation steps.”
Yeon Sik Jung, Professor, Department of Materials Science and Engineering, Korea Advanced Institute of Science and Technology
The scientists used deep learning, an artificial intelligence technology that can hierarchically retrieve particular elements of spectral information to classify data, to filter through the noisy signals. Their approach, called the dual-branch wide-kernel network (DualWKNet), was created to understand the link between spectral properties efficiently. According to Professor Jo, this capacity is essential for interpreting one-dimensional spectral information.
Explaining that DualWKNet enabled the team to find significant peaks in each class that was nearly indistinguishable in individual spectra, improving classification accuracies, Professor Jo states, “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, high classification accuracies of bacterial types and their media were achieved.”
“Ultimately, with the use of DualWKNet replacing the bacteria and media separation steps, our method dramatically reduces analysis time,” he added.
The researchers intend to use their framework to investigate more bacteria and media kinds, employing the data to create a training data library of numerous bacterial varieties in additional media to reduce the time required to gather and detect new samples.
“We developed a meaningful universal platform for rapid bacterial detection with the collaboration between SERS and deep learning,” Professor Jo stated. “We hope to extend the use of our deep learning-based SERS analysis platform to detect numerous types of bacteria in additional media that are important for food or clinical analysis, such as blood,” he concluded.
Rho, E., et al. (2022) Separation-free bacterial identification in arbitrary media via deep neural network-based SERS analysis. Biosensors and Bioelectronics. doi.org/10.1016/j.bios.2022.113991.