Diagnosis of melanoma by imaging mass spectrometry: Development and validation of a melanoma prediction model

In this joint work we partnered with Frontier Diagnostics, Vanderbilt and several board certified pathologists to develop a classification model capable of distinguishing melanoma from benign nevus tissue, using a MALDI-imaging based proteomics assay.

This assay was established based on samples of unambiguous nevi and melanomas. We used a histology-guided MALDI imaging based approach, in which expert pathologists first annotated regions of interest in the samples. We then acquired mass spectra within those ROIs, rather than doing full tissue imaging. The resulting spectra were then processed and fed into a classification model we developed.

Illustration of an example analysis of melanoma and benign nevus analyzed using MALDI imaging. Panels A-D highlight the microscopy at various zoom levels, whereas panel E shows the spectral content per tissue type. Taken from the publication with permission.

Our approach obtained a sensitivity of and specificity of on spot level predictions, and and sensitivity and specificity on the sample level, after aggregating spot predictions. These results place the assay among the top contenders for melanoma classification. Despite the admirable results, we are confident we can improve the assay further by incorporating data-driven quality control and doing more extensive machine learning modelling.

In this collaboration, Aspect Analytics was responsible for the full data analysis, starting at data preprocessing up to building and evaluating the classification models to perform tissue typing. In addition to this bioinformatics work, we have developed a full software backbone to support the workflow used by Frontier Diagnostics. You can find more details about the software side in our twin publication using the button below.

If you are interested in developing spatial omics based diagnostic assays, make sure to contact us for any questions you may have related to data analysis and/or dedicated software solutions.

Publication details

Rami N. Al-Rohil¹, Jessica L. Moore², Nathan Heath Patterson^2,3, Sarah Nicholson², Nico Verbeeck⁴, Marc Claesen⁴, Jameelah Z. Muhammad², Richard M. Caprioli^2,3, Jeremy L. Norris.2³, Sara Kantrow⁵, Margaret Compton⁶, Jason Robbins⁵, Ahmed K. Alomari⁷. Diagnosis of melanoma by imaging mass spectrometry: Development and validation of a melanoma prediction model, Journal of Cutaneous Pathology 12:, 1455–1462, 2021