A non-invasive method for concurrent detection of early-stage women-specific cancers

Abstract

We integrated untargeted serum metabolomics using high-resolution mass spectrometry with data analysis using machine learning algorithms to accurately detect early stages of the women specific cancers of breast, endometrium, cervix, and ovary across diverse age-groups and ethnicities. A two-step approach was employed wherein cancer-positive samples were first identified as a group. A second multi-class algorithm then helped to distinguish between the individual cancers of the group. The approach yielded high detection sensitivity and specificity, highlighting its utility for the development of multi-cancer detection tests especially for early-stage cancers.

Figure 1

Ion chromatograms of representative samples from the normal control and the individual cancer groups. The total run time for the LC resolution was 14 min, with every sample run being alternated with a blank run. For the blank run involved injection of a 1:1 mixture of methanol and water. Comparatively, spectra in each case were following a trend with major changes seen from 200 to 600 m/z with the time ranging from 3 to 11 min.

Figure 2

Age-wise detection of detected metabolites. Figure provides a graphical representation of the number of metabolites detected across the individual age groups, for the normal control set as well as the individual cancer groups. The cumulative unique metabolites detected in normal control samples were 5895. While, endometrial, breast, cervical and ovarian cancer samples were found to have 5971, 5982, 6300 and 6336 respectively.

Figure 3

Data processing pipeline. The data preprocessing pipeline used to render the data amenable to AI modeling is depicted here (details are given in the text).

Figure 4

PLSDA plot distinguishes between the individual cancers and also the normal controls. Figure presents a PLSDA plot of the matrix of sample-specific metabolites versus metabolite intensity for normal controls and the individual women-specific cancer sets. The separation obtained between the individual groups is shown. The R² and Q² values obtained are given.

Figure 5

AI workflow for distinguishing BECO cancers from normal controls and its application. Panel A depicts the AI workflow employed to test the AI model for distinguishing between the women-specific cancer group (BECO) from the Normal controls. Panel B depicts the results from testing of the trained model for distinguishing women-specific cancers (BECO) from normal controls showing clear separation of disease. The separation achieved between the cancer and the control group is shown in the form of a confusion matrix, with the resulting sensitivity and specificity values also given.

Figure 6

Partitioning of training and test data sets for the multiclass AI model. (A) shows the segregation of the individual cancer sets for training and testing of the multiclass AI model-2 (see text) for distinguishing between the individual cancers of the BECO group.

Figure 7

Testing the multiclass model for its ability to distinguish the individual cancer groups. Panel (A) shows the results of specifically testing the multiclass trained model for separation of endometrial cancer samples from the other cancers (breast, cervical, ovarian) based on model’s Endometrial scores. The resulting confusion matrix on applying a threshold shows good accuracy, sensitivity and specificity. Panel (B) shows the results of specifically testing the multiclass trained model for separation of breast cancer samples from the other cancers (endometrial, cervical, and ovarian) based on model’s Breast scores. The resulting confusion matrix on applying a threshold shows good accuracy, sensitivity and specificity. Panel (C) the results of specifically testing the multiclass trained model for separation of cervical cancer samples from the other cancers (breast, endometrial, ovarian) based on model’s Cervical scores. The resulting confusion matrix on applying a threshold shows good accuracy, sensitivity and specificity. Panel (D) shows the results of specifically testing the multiclass trained model for separation of ovarian cancer samples from the other cancers (breast, endometrial, cervical) based on model’s Ovarian scores. The resulting confusion matrix on applying a threshold shows high accuracy, sensitivity and specificity.

Figure 8

Preparation and scheduling of QC and samples for UHPLC-MS/MS. A small aliquot of each sample (coloured cylinders) was pooled to create a QC sample (multi-coloured cylinder), which was then injected periodically (every 50th injection) throughout the batch run. Variability among consistently detected metabolites was used to estimate overall process and batch variability. Every sample injection was followed by a blank injection to prevent carryover between the sample runs.