Using machine learning of 17 genomic signatures, researchers from IIIT Delhi and CSIR-IMTECH have developed a webserver called CancerSPP that will help clinicians in classifying cutaneous melanoma as either primary or metastatic. The accuracy is over 89%, and both sensitivity and specificity are high.
Using the expression of 17 key genes (messenger RNAs) it is now possible to distinguish primary and metastatic cutaneous melanoma, which is the most common type of skin cancer. While 11 of the 17 genes have already been reported by other studies for cutaneous melanoma, it is for the first time that the potential role of remaining six genomic signatures in classifying samples as either primary or metastatic skin cutaneous cancer has been made.
The 17 genomic signatures, which were identified by a team led by Prof. Gajendra P.S. Raghava from the Indraprastha Institute of Information Technology (IIIT), New Delhi, have high accuracy — over 89% — in discriminating metastatic from primary skin melanoma. These signatures also have high sensitivity (in case tumour is metastatic), and high specificity (in case the tumour is primary). The results were published in the journal Scientific Reports.
Unlike in the case of primary skin melanoma, people with metastatic cutaneous melanoma have reduced survival rate and higher mortality rates. It therefore becomes important to be able to identify and classify skin cutaneous melanoma as either primary or metastatic so correct therapeutic strategies can be chalked out and survival rates improved in patients.
Six machine learning models were used to study and validate the genomic signatures. They used expression profile of messenger RNA, micro RNA and methylation profile for discriminating tumour as primary or metastatic. “We found the messenger RNA expression profile was the strongest predictor of metastasis. The mRNA expression profile performed better than micro RNA and methylation profile of the patients,” says Harpreet Kaur from Institute of Microbial Technology (CSIR-IMTECH), Chandigarh and one of the first authors of the paper. “Of the six models used, one (SVC-W) model showed better ability to discriminate metastatic from primary tumours of validation dataset with overall accuracy of over 89%.”
While, messenger RNA outperformed microRNA in discriminating the status of the tumour, a particular microRNA was found to be a “strong predictor of metastatic melanoma”.
Besides helping in distinguishing the kind of melanoma, the genomic signatures can also help in further categorising different stages of metastasis. For instance, it can tell if the tumour has spread or not to the lymphatic nodes, which is an early stage of metastasis. Likewise, it can tell if the cancer has spread to distant parts of the body, which is a late stage of metastasis.
Six machine learning models were tested and used for classifying the tumour as either primary or metastatic. Of the six models, one model — Support Vector Classification with Weight (SVC-W) — has an accuracy of nearly 89.5%.
The researchers have further integrated the major prediction models in the webserver called CancerSPP that will help clinicians in classifying cutaneous melanoma as either primary or metastatic using RNA sequence data, microRNA and methylation expression data. “It will also help in knowing the different states of metastatic samples,” says Kaur. “The analysis module in the CancerSPP webserver will provide information on the role of each of the important genes in various stages of metastasis and whether the expression of a gene is up-regulated or down-regulated.”