New analytical tools for biomedical applications using machine learning and spectroscopy

Machine Learning

Evaluation and monitoring of bacterial growth and response to antibiotics is of great importance not only in the laboratory but also in food, environmental, medical and pharmaceutical applications. James Chapman of RMIT University, Melbourne, Australia, and colleagues demonstrate the power of near-infrared (NIR) and ultraviolet-visible (UV-Vis) spectrophotometry combined with machine learning tools (such as partial least squares discriminant analysis) has been actively researched. To examine, measure and monitor bacterial growth stages during treatment with antibiotics.Chapman spoke Spectroscopy about his efforts.

A recently published paper utilizing near-infrared spectroscopy (1) states that the measurement of the bacterial growth phase using optical density at 600 nm (OD600) is the “gold standard” for industry and research institutions. says. Why do you believe this to be the case, and have you discovered that this practice may be challenged?

Using the OD600 represents the most rapid single measurement in many pathology, microbiology and biochemistry laboratories and is routinely used to determine the number of cells in solution. This measurement is based on the established science of using turbidity and Beer-Lambert’s law as measures of cells grown in solution to obtain an estimate of the concentration of cells per unit volume. This his OD600 measurement can be done with a multi-wavelength or single-wavelength spectrophotometer. As you know, these instruments are abundant in most laboratories.

Our discovery was made while using a multi-wavelength spectrophotometer. We therefore hypothesized that common biochemical information is missing across the spectrum, or missing altogether. We thought that important information could be obtained from the spectra. For example, DNA absorbs at 260 nm in the UV-visible region and RNA absorbs at 280 nm, among other important biochemicals. We perceive these puzzle pieces as additional information, as well as cell numbers. This multi-factor information enhances our ability to assign a fingerprint of what is happening within the system. As a result, spectrophotometers can be used to understand the interaction of microorganisms with antimicrobial agents, and all interactions within complex matrices.

A second published paper (2) explored the feasibility of UV-Vis spectrophotometry to examine the growth stages of bacteria upon treatment with antibiotics and determined resistance points when antibiotics were introduced into the culture medium. I showed what I could understand, which improved my understanding. Study of biochemical changes in infectious agents. How do you compare NIR and UV-vis as optimal methods for assessing bacterial counts, and how do these two methods compare to other available analytical techniques in terms of speed and accuracy? compared to ?

There are two reasons for using these two methods. One is that NIR can measure samples with higher energy transmittance, and the other is that UV-visible spectrophotometry offers a multi-parameter method to see more than standard optical density methods. . Both are very different methods, but when combined they provide very powerful information for antimicrobial resistance applications. For example, NIR can be used to scan samples through sample vials, so not much sample preparation is required. All major methods can be used when time is of the essence.

The spectroscopy we use is a rapid, non-destructive and environmentally friendly technique. This offers the advantage of less sample preparation, less sample destruction, and near-instantaneous measurement of the sample. High-throughput sampling can also be performed using our developed method. It is not surprising when comparing these to well-established, highly qualitative and quantitative techniques such as GC-MS, NMR and LC-MS. Some techniques require lower limits of detection, but spectrophotometry is the method of choice when information is needed in a rapid manner. For example, it currently takes up to a week to diagnose and treat antibiotic contamination. Pushing the time boundaries using the method we developed means that we can consider reducing the time frame of antibacterial diagnostics.

How do you distinguish between using terms artificial intelligence and machine learningDo you think artificial intelligence could be useful for this application?

Artificial intelligence is an umbrella term that includes machine learning. Therefore, any use of machine learning is a form of AI. AI is essential for all applications of this kind. This will allow us to revolutionize our methods and generate faster responses to diagnoses. We feel these are essential to the design of new analytical systems in the future. In addition, automating the data analysis process creates better AI models that can help you understand and characterize unknown systems much faster.

How is this job different from what you or others have done in the past?

Our group’s research is state-of-the-art and some of the first reported in this field. Examining multiparameter systems in complex matrices is unheard of when analytical methods usually aim to extract the analytes of interest first, which can be used, for example, in solid-phase extraction, sample clean-up , easily achievable through chromatography. However, our method extracts this information in one step, even when measuring antibiotics, microbes, and the nutrient broth in which they are measured in a complex matrix.

Please summarize your findings.

We combine chemometrics and machine learning algorithms to develop optimal diagnostic, therapeutic, and decontamination strategies. Our initial method showed that adding various concentrations of broad-spectrum antibiotics to the growth medium produced unique changes in the UV-Vis and NIR spectra. Applying preprocessing techniques to the data will allow us to obtain more information for monitoring and differentiating treatments.

Were there any limitations or challenges you faced at work?

As with all experiments and scientific work, there are always challenges. We are working on improving our detection limits and trying to train our AI system to improve the quantification of the models we create. The ability to sort, process, and analyze data is still very manual. However, our group and PhD students are enthusiastic about developing automation. The types of projects we work on are mostly interdisciplinary and require multidisciplinary teams to solve these problems. As such, the Chapman Group includes mathematicians, chemical analysts, materials scientists, computer scientists, biologists, and chemists, all passionate about solving these challenges.

Can you summarize the feedback you’ve received from others on this piece?

We have received a lot of interest in this initiative. Industry is keen to partner with us, and we now have many collaborators looking to expand our application base, such as cancer detection in tissue biopsies (looking for key biochemical markers). exist internationally. We are always looking for new collaborations and challenges to tackle. So this network base is pretty dynamic and everyone is welcome to join us.

What are the next steps for this research?

We are already working on several steps in this study, including improved limits of detection, multiple multi-microbial samples (which better reflect real-world problems), and a variety of newly developed antimicrobial agents. It will be published soon, focusing on the use of antimicrobial agents. . We are a young and dynamic team and the development of new antimicrobial solutions is also a separate department within the group, so the characterization of such systems and interactions is at the forefront of our activity. We have automated many steps and are now generating deep learning methods. In the next phase, we will utilize a hyperspectral system to perform these measurements. All with spectra, all analyzed with chemometrics, all automated with AI. Really, really exciting.


(1) Truong, VK; Chapman, J. Cozzolino, D. Monitoring bacterial response to antibiotics and temporal growth using near-infrared spectroscopy combined with machine learning. Food analysis method 2021, 14, 1394-1401. DOIs:

(2) Chapman. J.; Orel Trigg, R. Kuun, Kentucky.Truong, VK; Cozzolino, D. A high-throughput machine learning resistance monitoring system for determining resistance points. Escherichia coli In combination with tetracycline: a combination of UV-visible spectrophotometry and principal component analysis. biotechnology. bioengineering. 2021, 118 (4), 1511-1519. DOIs:

About the interviewee

Source link

Leave a Reply

Your email address will not be published. Required fields are marked *