Detection of Airborne Biological Particles in Indoor Air Using a Real-Time Advanced Morphological Parameter UV-LIF Spectrometer and Gradient Boosting Ensemble Decision Tree Classifiers

Round 1

Reviewer 1 Report

The manuscript describes a new classification approach to enable the identification and quantification of primary biological aerosol particles (PBAPs) more precisely at high time resolution by using a MBS UV-LIF（Multiparameter Bioaerosol Spectrometer Ultraviolet Laser-induced Fluorescence). Five parameters based on the 2D profiles of aerosol particles were introduced, with which categorizing some new group of particles such as cotton fibers turned to be available. The authors investigated the efficiency of the new classification with laboratory-prepared particles, and further used the approach to particles in a multifunctional building and got some interesting results. The manuscript contains many contents worthy of the publication in the journal. However, there are many vague points in the manuscript.
I have two major concerns. One is what is the focus of this manuscript: the efficiency and effectiveness of the approach or the results from the application in the building (or both). The title of the manuscript sounds like the later one while the abstract the former one, while the text of the manuscript waves in-between, leaving both points incomplete. Another is that the manuscript was not carefully prepared. Many details which are essential for understanding correctly the contents are missing. At the same time, there are many tedious and repeat descriptions. Some of the points are raised here and authors are requested to carefully revise the whole manuscript.
1. Title and the contents of the abstract should be identical.
2. Abstract: make the contents accurate and easily understood.
3. In Line 50-58, the authors used many sentences to show the harmfulness of asthma, which are far away from quantification and identification of PBAP, the topic of this study.
4. In Section2.1, this section is crucial for helping readers to understand the theoretical fundamental of this work, but it is hard to follow. Descriptions need to be expressed in simple and easily understandable ways, such as Lines 140-143 and 148-151. Long sentences should be avoided. The abbreviation of CMOS in Line 154 is mentioned first time. A full name of CMOS should be given, and a brief introduction of CMOS is necessary. In addition, simply describing the advantages and disadvantages of the 5 parameters (in Line 165) will be beneficial to understanding the merits of the methodology extended in this study.
5. In 3.1 section, reasons for why these five unwashed bacterial species were selected are necessary.
6. Details of sample processing for Table 1 are necessary. SEM images of particles seem helpful.
7. The fonts of axis labels and legends in all the figures are too small and not clear. Please improve it. The microbial name in the caption of Fig.1 should be consistent with manuscript in italic type. Please explain Y-axis label in Fig.2. In addition, Table.2 should be re-formed.
8. Line 376, the authors presented that the cotton fibers are heterogeneous. However, a cotton t-shirt was used to generated cotton-like particles. Why the cotton fibers are not homogeneous?
9. Line 417, how do the fungal material and chemical process influence the fluorescent and morphological characteristics? Some interpretation or discussion are needed.
10. Line 464, the caption of the Fig.7 is not complete.
11. In Line 476, it seems that not only both fungal and cotton-like aerosol display the midday maxima trend, also the bacteria-like and unclassified aerosol. The interpretation is not adequate to clarify these uncertainties.
12. The conclusion part is very tedious and needs large and proper simplification.
13. A list of abbreviation may be helpful.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

General comments

 

   This research reported on the monitoring techniques of airborne biological particles using fluorescent detection systems (UV-LEF). The authors find that the classification against training classes are available for the identification among fungi, bacteria and so on. The establish of such biological-particle monitoring is very useful for the proceeding of biological aerosol researches. However, I think that this manuscript contains redundancy parts and need to explain about the training classes more clearly (because of this importance for this paper). Additionally, morphological characteristics of microbial particles should be discussed using the real morphological forms of fungal cells.

    I recommend this manuscript is resubmitted after total redraft or rejected.

 

Some major comments:

 

1. As described, there are some redundancy parts which cause confusion for readers. The aims of study should be described more clearly instead of the section from L76 to L133. Other section also should be improved to be clearer for indicating the algorithm processes. I think almost readers of this journal Atmosphere are not familiar with this algorithm and this device.

 

2. This detection system targets the metabolic products of biological particles. However, almost biological particles are expected to be dead or un-activated in atmosphere due to environmental stressors. Why can the fluorescence of biological particle be detected? Additionally, these metabolic products are thought to be placed biasedly in cells (in particular eukaryote). I think that the morphological characteristics is not so important for this detection. The authors should explain about morphological characteristics of microbial particles using the real morphological forms of fungal cells (spores, filament).

 

3. Although biological particles are monitored outside using the established UV-LEF, there are no evidence demonstrating that the detected signals are originated from bacteria or fungi. Other methods, such as microscopic observation or real-time PCR, are needed as the evidence. If not possible, this point have to be discussed.

 

Some minor comments:

L20: “Hellinger distance metric” should be explained shortly here.

L24-27: Fungal cells of single species include spore and filament types. Did the difference influence this detection?

L76-133: This part is thought to be important for describing about the originality of this study, but the author have to summarize here to focus on the merits of UV-LIF and machine learning techniques. Finally, the original points among machine learning techniques for this study should be indicated and the clear explanation (a few sentences) about this study aims is required.

L107: The word “Supervised” is thought to be a conceptional word. Can the authors use more concrete word that indicate the original technique for this study?

L126: The theory of Gradient boosting ensemble decision trees (GBA) should be explained clearly in Materials and methods.

L166-176: I think figure is needed for describing these factors.

L186-239: This section is important for this study, but it is too long and hardly understood. I recommend the authors redraft here to summarize original points of the research approach and improve to be more understandable explanation.

L252: How to prepare these powders? This information is required for evaluating the influences of morphological characteristics on fluorescence detection. Moreover the forms of each microorganisms should be described.

L279: These fungi sometimes form black color spores. Can they be detected by aut-fluorescence detection?

L292-357: I think this section is too long and can be shorten for indicating the fluorescence index of each biological particle.

L299: Does “is not unexpected” mean  “is expected”?

L303: “in June to September” can be changed to  “from June to September”.

L339: The explanation about Hellinger distance metric is needed before this parts.

L358-425: This section is thought to be also too long and should focus just on the explanation for fluorescence detection matching to each biological categories.

L431-499: Support data or discussion is needed for describing the evidence for the presence of airborne microorganism outdoor.

Figs: Totally letter size is small. Graphs should be modified for clear viewing.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

In this manuscript, the authors have reported the results from a study evaluating the utility of supervised machine learning to classify single-particle ultraviolet laser-induced fluorescence signatures to investigate airborne PBAP concentrations in indoor air. They applied a Multiparameter Bioaerosol Spectrometer for this study, indicating good utility of this approach for classifying airborne bacteria and fungi. The manuscript has presented original data obtained by the novel approach to quick and real-time detection of airborne microorganisms. Besides, the authors have revealed advantages and some limitations of their technique. Generally, the study has well been prepared, conducted and described by the authors. However, some items require their attention to improve the manuscript.

1. The introduction might end with clearly described objectives of the study. Consequently, the conclusions might correspond to the aims. As it stands, the conclusions are vague and too long. They should avoid describing the study again in brief. Also, the results have not to be repeated in this section. However, the novelty of the approach might be underlined as a conclusion.
2. Citing references in the text, they should use only respective numbers, e.g. Khan and Karuppayil [15]; correct all items in the manuscript.
3. Aspergillus, Cladosporium and Penicillium are genera, not species; thus, the sentence might be corrected (l. 66-68) as “species belonging to Aspergillus, Cladosporium and Penicillium…”. Otherwise, writing of species the names can be used as Aspergillus sp., Cladosporium sp. and Penicillium sp.
4. For the figures, larger fonts might be used for the description of axes and legends. As it stands, a reader can hardly discern these descriptions.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The authors have well revised the manuscript in response to my major concerns, and its readership was largely upgraded.

Author Response

We again thank the referee for their helpful comments and suggestions throughout the review phase to improve the revised manuscript. 

Reviewer 2 Report

 This manuscript has been revised in dependence on reviewer’s comments. The revised version has been improved to be clearer and sophisticated. However, there are some issues for Table and Figures still.

    I recommend this paper is accepted in this journal after minor revision.

 

Table 2: I think this style is not normal style as a table of research paper. The under line parts, such as Bacteria… and Predicted label, can be moved to upper line parts. The “True label” and “Predicted label” have to be defined in table.

 

Figures: The letters of all figures are thought to be small still and hardly recognized by readers.

Author Response

This manuscript has been revised in dependence on reviewer’s comments. The revised version has been improved to be clearer and sophisticated. However, there are some issues for Table and Figures still.

I recommend this paper is accepted in this journal after minor revision.

We thank the referee for their helpful comments and suggestions throughout the review phase to improve the revised manuscript.

Table 2: I think this style is not normal style as a table of research paper. The under line parts, such as Bacteria… and Predicted label, can be moved to upper line parts. The “True label” and “Predicted label” have to be defined in table.

Table 2 is in the format commonly used for confusion matrix tables, however, we have adjusted the table to reflect the standard journal table format.  Further formatting to tidy the table will be performed by the journal typesetters when the manuscript is converted to LaTeX.  We now also include definitions of the labels in the table caption as requested.

Figures: The letters of all figures are thought to be small still and hardly recognized by readers.

We have further increased all figure font sizes as much as was possible without causing other figure formatting issues.