Grantee Research Project Results

Final Report: PCR-Free Environmental Waterborne Bacteria Detection Using Raman Spectroscopy and Deep Learning

EPA Grant Number: SU840575
Title: PCR-Free Environmental Waterborne Bacteria Detection Using Raman Spectroscopy and Deep Learning
Investigators: Li, Yiyan
Institution: Fort Lewis College
EPA Project Officer: Brooks, Donald
Phase: I
Project Period: August 1, 2023 through July 31, 2024
Project Amount: $25,000
RFA: 19th Annual P3 Awards: A National Student Design Competition Focusing on People, Prosperity and the Planet Request for Applications (RFA) (2022) RFA Text |  Recipients Lists
Research Category: P3 Awards

Objective:

The goals of this proposed project are: 1) To create a deep learning model and a Raman spectroscopy workstation for the rapid detection of environmental waterborne bacteria in the laboratory. 2) To utilize the Raman station for testing water samples from rivers in Colorado. 3) To engage with communities, including Native American tribes in the Four Corners region, about the significance of water resource protection and the technologies available for environmental conservation.

Summary/Accomplishments (Outputs/Outcomes):

The Polymerase Chain Reaction (PCR) technique has been used as the gold standard of amplifying target DNAs to detect microorganisms or viruses from environmental water samples. PCR requires a power-consuming thermocycler, a set of perishable chemical reagents, standard wet lab supplies, and benchtop imaging and screening equipment. Following a 10-hour culturing process, the sample preparation, thermocycling, and imaging procedures of a few PCR trials take at least 3 hours in a lab by trained personnel. Here, we propose a new method that uses Raman spectroscopy and deep learning to identify the bacteria genotypes. The principle behind the technology is the Raman effect, where a sample exposed to a laser will scatter light at a frequency different from that of the incident light (these are known as vibrational signatures). The process of generating spectra is considerably simpler and faster than standard PCR and allows for higher throughput, fewer preparation materials (which risk error), and non-specialists to gather high-quality data. A deep learning model is trained by thousands of documented bacteria samples in advance and is able to identify new or unknown bacteria samples from environmental water samples. The turnaround time of the post-culturing procedure of a single Raman test is only 1-2 minutes compared to the counterpart PCR trial which takes at least 3 hours.
Summary of Findings:
1. The team established a bacteria concentration protocol using the Innovaprep pipetting system, achieving a 93% recovery rate at a concentration of 1142 CFU/mL and a 20.8% recovery rate at 114.2 CFU/mL.
2. This study utilizes a newly released dataset of Raman spectra from bacterial samples, along with an enhanced Convolutional Neural Network (CNN). The results show improved identification accuracy and reduced network complexity. The AI model reaches an identification accuracy of around 86%, with near real-time identification speeds.
3. The team developed a micro-Raman data computing system and explored various smart, portable platforms for edge AI computing. The AI model was successfully implemented on the Raspberry Pi 5 platform, which achieved similar accuracy to a laptop, producing a confusion matrix with 83% classification accuracy. Adding a touch display transformed the system into a handheld, user-friendly device for real-time AI-based detection.
4. E. coli, E. cloacae, and S. epidermidis samples were detected using the Horiba LabRAM HR Evolution Spectrometer and our CNN AI model.
5. The team also used traditional PCR techniques to assess potential human fecal contamination in the Weminuche Wilderness, but no positive Bacteroides PCR results were found in the water samples; The team collected samples from the concerned spots of Animas River reported by the Mountain Studies Institute and analyzed the samples using chromogenic enzyme substrates. No signs of contamination over the recreational water limits of E. coli.

Conclusions:

This study successfully advanced the detection of environmental waterborne bacteria by integrating innovative technologies and methodologies. The development of a bacteria concentration protocol using the Innovaprep pipetting system achieved a high recovery rate, demonstrating its effectiveness for waterborne bacterial sample concentration. The use of a newly released Raman spectra dataset, combined with an enhanced Convolutional Neural Network (CNN), resulted in an identification accuracy of 86%, with near real-time processing speeds, reflecting the potential of AI in environmental monitoring. The team also demonstrated the feasibility of deploying the AI model on a portable Raspberry Pi 5 platform, maintaining classification accuracy of 83% and enhancing the system’s accessibility through a handheld touch display. Detection of key bacteria, including E. coli, E. cloacae, and S. epidermidis, was achieved using the Horiba LabRAM HR Evolution Spectrometer, showcasing the system’s precision. Additionally, traditional PCR methods were employed to assess fecal contamination in the Weminuche Wilderness, revealing no positive results, while water samples from the Animas River confirmed the absence of contamination above recreational water safety limits. Overall, this study highlights the potential of combining AI, Raman spectroscopy, and portable devices to provide reliable, real-time water quality monitoring, offering valuable insights for environmental protection efforts.

References:

[1] M. L. Devane, E. Moriarty, L. Weaver, A. Cookson, and B. Gilpin, "Fecal indicator bacteria from environmental sources; strategies for identification to improve water quality monitoring," Water Research, vol. 185, p. 116204, 2020.
[2] D. A. Holcomb and J. R. Stewart, "Microbial indicators of fecal pollution: recent progress and challenges in assessing water quality," Current environmental health reports, vol. 7, pp. 311-324, 2020.
[3] K. L. Offenbaume, E. Bertone, and R. A. Stewart, "Monitoring approaches for faecal indicator bacteria in water: Visioning a remote real-time sensor for e. coli and enterococci," Water, vol. 12, p. 2591, 2020.
[4] D. Moschou and A. Tserepi, "The lab-on-PCB approach: tackling the μTAS commercial upscaling bottleneck," Lab on a Chip, vol. 17, pp. 1388-1405, 2017.
[5] M. Maurin, "Real-time PCR as a diagnostic tool for bacterial diseases," Expert review of molecular diagnostics, vol. 12, pp. 731-754, 2012.
[6] P. Rules, "Clean Water Act Methods Update Rule for the Analysis of Effluent," Federal Register, vol. 80, 2015.
[7] E. van Pelt-Verkuil, A. Van Belkum, and J. P. Hays, Principles and technical aspects of PCR amplification: Springer Science & Business Media, 2008.

Supplemental Keywords:

Deep Learning, Raman Spectroscopy, Bacteria Detection, Water Quality Monitoring