Identifying Filamentous Bacteria? There's an App for That.

Novozymes offers a fast and simple way for clean-water plants to troubleshoot problems caused by filamentous bacteria.

Identifying Filamentous Bacteria? There's an App for That.

Plant Assistant reports on microscopic evaluation of filaments can be displayed on desktop computers or mobile devices.

Interested in Laboratory?

Get Laboratory articles, news and videos right in your inbox! Sign up now.

Laboratory + Get Alerts

Excessive filamentous bacteria can cause severe disruption in clean-water plants, most notably poor settling in final clarifiers.

When a filament issue arises, it’s essential to find the cause, and that starts with a microscopic study to determine the abundance and species of filaments present. Some plants have the equipment and expertise in-house to make the determination. Some do not and therefore need to consult with experts.

Traditionally, that means sending samples of activated sludge to a laboratory for examination, a process that takes time and delays resolution of the problem. Now, believe it or not, there’s an app for that. Novozymes, a company that offers enzyme and microbiological solutions to clean-water plants, as well as biofuel producers, bakeries and breweries, has developed Plant Assistant, a web-based application that can identify and score common wastewater filaments from a microscopic picture.

The company’s data scientists and biological experts have used thousands of images of filaments to train the algorithm to identify and differentiate specific characteristics. At present, the app can identify eight filaments with more than 90% certainty, and plans are underway to expand that capability. Anders Klitholm Jessen, manager of digital transformation, and Matthew Livingston, global marketing manager, talked about the app in an interview with Treatment Plant Operator.

TPO: How did your company develop the capability to create this tool?

Livingston: For a long time, we have developed expertise around doing microexams. We have people on staff who are adept at identifying filaments, understanding their morphology and doing the right staining techniques. One advantage we have is that because we have visited a number of treatment plants over a long time, we have a good understanding of how to do these examinations.

TPO: How did you get the idea to automate the identification process?

Livingston: In brainstorming around how we could use digital tools, we noted that filament identification is something that’s labor intensive and requires a certain skill set. So we asked our information technology specialists if this was something that could be automated, and they tended to think it was possible.

TPO: How did you go about training the algorithm?

Klitholm Jessen: Through our many contacts in the industry, we have received samples from municipal and industrial facilities that are experiencing different issues. Our experts have used these samples to show different images in order to teach the algorithm. The training is based on about 100 to 150 images per filament. That’s a little over 1,000 images that we have in our initial dataset. We are now looking to obtain more images to make the algorithm even better. We plan to expand the scope so we can identify up to 20 filamentous organisms and potentially also higher life forms, floc particles and other microorganisms.

TPO: About how many types of filamentous bacteria do clean-water operators commonly see in their treatment processes?

Livingston: About 20 are considered the most prevalent, according to reference manuals. From our experience, about 10 of those are the most common. As you get into the teens, you see them less frequently; then once in a while organisms pop up that are not very common at all.

TPO: What distinguishing features do these filaments have that your app can recognize?

Livingston: There are different sizes. Some are straight; some are curved; some are bent. Some are skinnier or fatter. The cell shapes are different — some are square, some rectangular, some sausage shaped. Some filaments are smooth and have a casing or sheath on the outside. Some have bacteria growing perpendicular to the filament, so it looks like a hairy filament. Some features are easy to tell apart, and some are more difficult.

TPO: Can you train an algorithm to recognize even the smaller differences?

Klitholm Jessen: We are able to identify most of the filaments at a normal magnification of 400x, although some require a larger magnification. In principle, a well-trained expert can distinguish filament A from filament B. A machine is also able to do it via machine learning. With enough data, the algorithm can be trained to recognize the different filaments with reasonable accuracy.

TPO: When a filament problem arises at a treatment plant, is it typically just one filament, or can there be multiple filaments in the mix?

Livingston: It can be either. The benefit of identifying the filament is that it enables operators to understand the cause. Most filaments are associated with specific substrates such as organic acids, sulfides, or fats, oils and grease (FOG) or with certain parameters such as low dissolved oxygen, low food-microorganism ratio, or a deficiency of nitrogen or phosphorous. If one driving condition is causing the filaments, you may see multiple filaments within that category. Sometimes there are multiple causes, and you may identify multiple filament types.

TPO: What equipment does a plant need in order to use the app?

Klitholm Jessen: The two critical things a plant needs are a phase-contrast microscope and a device on which to run the app. That can be a desktop or laptop computer, tablet or smartphone.

TPO: Do users receive any guidance on how to create the images for analysis?

Livingston: If someone can take a high-quality picture at 400x, then we would be able to use that as the preliminary round. You can assess the abundance of filaments using a light microscope, but to identify the filaments, you really need a phase-contrast scope.

TPO: Once operators acquire the necessary image, how do they use the app?

Klitholm Jessen: They upload the image to the app. Our algorithm analyzes the image and returns a list of potential matches. It lists each filament with a match accuracy based on the image: for example, Filament A, 78%; Filament B, 29%; and so on. The app includes a reference library that they can use to check the results against if they’re not fully satisfied with the answer.

TPO: Once the image is uploaded, how long does it take to get the answer?

Klitholm Jessen: It takes between half a second and three seconds. If you compare that with the traditional way of mailing samples and getting results back from experts, it’s fair to say there is quite a substantial improvement. 

TPO: How do users pay for this service?

Livingston: We are not looking at this as a pay service. We see it as something that can be useful throughout the industry. It’s a way to help people who have problems confirm those issues and speed their path to a solution. Some filament issues can be addressed using technologies we have.

TPO: What are examples of problems your technologies can help solve?

Livingston: For filaments driven by low DO, we don’t have solutions that can help, but if we can help a plant identify the problem, that’s good for us and good for them. For filaments driven by FOG, that’s a problem we can solve using enzyme or microorganism formulations. In addition, sometimes there’s a course of action that’s not related to adding our product. Filament abundance can be such that we’ll recommend they chlorinate their system to kill some of the filaments, and then reseed the system to get their biomass back to a healthy condition. We make some of those types of recommendations.



Discussion

Comments on this site are submitted by users and are not endorsed by nor do they reflect the views or opinions of COLE Publishing, Inc. Comments are moderated before being posted.