A Rapid and Affordable Screening Tool for Early-Stage Ovarian Cancer Detection Based on MALDI-ToF MS of Blood Serum

Round 1

Reviewer 1 Report

In this retrospective study, the Authors have evaluated the feasibility of MALDI-ToF mass 20 spectrometry-based test for screening of ovarian cancer using 181 serum samples from patients. Interestingly, they identified a high 99% sensitivity and 92% specificity in this setting.

The early detection of ovarian cancer is an important tool in clinical practice, because often this neoplastic disease is silent, and at diagnosis it could appear as an advanced disease. I agree with the Authors when they stated that CA 125 and transvaginal ultrasound are not sufficient for early detection. So, new technologies are necessary in this way, as the purposed MALDI-ToF MS. Although this system is not easy to translate in clinical practice, I find it promising further for important clinical implications.

As minor comment, I would know if possible about the CA 125 levels in the 142 women with ovarian cancer.

Author Response

We thank reviewer 1 for their comments and to recognize the relevance of our work.

Regarding your suggestion, we recognize the relevance of contrasting the predictions of our models with the patients’ CA 125 levels. This would be of key importance to reinforce the soundness of our approach as a diagnostic tool. However, this data was not available during sample acquisition in this retrospective study. Further, we have processed the serum samples with CHCA and SA which may interfere with a posterior CA 125 quantification.   Nevertheless, we plan to compare the  CA 125 levels as predictors alongside with our MS pattern-based models in a future screening study that would serve as a validation of the clinical application of our approach.

Reviewer 2 Report

Pais et al. developed and applied the MALDI-ToF MS method in conjunction with bioinformatics pipeline to classify different stages of ovarian cancer using easily accessible biofluid, serum. Authors also show the sensitivity and specificity better than existing methods. Overall manuscript is well written and meticulously designed experiment to draw the conclusion. While the method is rapid and simple, main limitation of this study is the identification of molecular features that differentiate the stage of cancer. It is suggested to authors should put effort to get identification of molecular features. In addition, bioinformatic pipeline is not well described to implement in similar research questions by other researchers in the field. Other concerns include

• No patients’ characteristics data such as age, sex, smoking status etc. are available in manuscript.
• Page 2, line 71 “(MALDI-ToF MS) is a very sensitive, affordable and accurate technique” MALDI-ToF MS is expensive instrument, please comment!
• Page 4, line 143-144 “Replicates for each 143 patient’s sample were also generated to ensure high-quality assessment.” How many replicates were generated please specify?
• Page 4, line 166, “An improved peak finding algorithm was implemented in this pipeline to extract” please describe what is improved peak finding algorithm?
• Page 7, line 256, it is not clear how model was trained and validated to obtain binary classification data

Author Response

We thank the reviewer for their comment, below are the point-by-point answers.

1.  “main limitation of this study is the identification of molecular features that differentiate the stage of cancer. It is suggested to authors should put effort to get identification of molecular features”

Although we agree that finding the molecular markers that compose our identified MS patterns would provide additional biological support, this is out of the scope of our work. Our focus was to demonstrate that patterns of complex mixtures of molecular entities (glycoproteins and metabolites with distinct degrees of chemical modifications) in the serum have higher predictive power in comparison to individual markers, and these can be captured by MALDI-ToF mass spectrometry regardless of their molecular identification using scoring-based mathematical models. The underlying assumptions and rationale for our methodology have been discussed in previously published work (see refs 17-22 , 31 and 32) referenced in lines throughout the text of the manuscript (see lines 87-99, 125-130 and 422-424). Further, the identification of the molecular markers of our identified patterns is a methodological challenge that requires coupling with HPLC to isolate each component which we intend to inspect in a future screening study. 

1. “bioinformatic pipeline is not well described to implement in similar research questions by other researchers in the field”

A detailed description of the bioinformatics pipeline is available in Pais et al 2019 (ref 18) and the machine learning-based scoring approach described in Pais et al 2020 (ref 20). In the work that is being reviewed, we focus on describing the adaptations made on these methodological approaches for addressing the ovarian cancer pattern identification on women serum samples (see methods section). We believe that researchers in bioinformatics can easily implement similar methodologies based on the combination of our previous published work (ref 18 and 20) with the currently presented work.

1. “Page 2, line 71 “(MALDI-ToF MS) is a very sensitive, affordable and accurate technique” MALDI-ToF MS is expensive instrument, please comment!”

We referred to MALDI-ToF MS as an affordable technique/solution for the development of diagnostic tests/platforms based on its reduced operational costs and reagents/consumables in comparison with other high-throughput technologies (e.g NGS and RNAset) and biomarker specific immunoassays. This has been already discussed in specific reviews (see refs 11 and 14). Although it is true that acquiring a MALDI-ToF machine is expensive, the savings with operational costs of the tests can economically compensate for the initial investment in an reasonable time span. We further discussed this issue for the application to the early detection of ovarian cancer (see lines 421-457). 

1. “Page 4, line 143-144 “Replicates for each 143 patient’s sample were also generated to ensure high-quality assessment.” How many replicates were generated please specify?”

We improved the manuscript text to clarify the replicates. Please visit the revised version (lines 143-145 and lines 158-160). 

1. “Page 4, line 166, “An improved peak finding algorithm was implemented in this pipeline to extract” please describe what is improved peak finding algorithm?”

We improved the manuscript text to clarify the peak finding algorithm. Please visit the revised version (lines 169-178). 

1. “Page 7, line 256, it is not clear how model was trained and validated to obtain binary classification data”

All models were trained using a random selection of 50% of the data and validated using 100% of the available dataset.  We used a novel state-of-the-art ML algorithm EvA-3 for pattern identification with predictive power. The methodology is described in the Methods section (lines 206-217). 

1. “No patients’ characteristics data such as age, sex, smoking status etc. are available in manuscript”

We add an additional supplementary file “ID logbook.csv” containing patient characteristics.