Towards a potential pan-cancer prognostic signature for gene expression based on probesets and ensemble machine learning

Our plain-English summary of Towards a potential pan-cancer prognostic signature for gene expression based on probesets and ensemble machine learning, by D. Chicco, A. Alameer, S. Rahmati, et al (BioData Mining, 2022).

The study

• Authors: D. Chicco, A. Alameer, S. Rahmati, G. Jurman
• Journal: BioData Mining
• Published: 2022;15(1):28
• Cited by: 3 articles, as counted by NIH iCite
• Read it: PubMed · Publisher · Free full text

Abstract

Cancer is one of the leading causes of death worldwide and can be caused by environmental aspects (for example, exposure to asbestos), by human behavior (such as smoking), or by genetic factors. To understand which genes might be involved in patients' survival, researchers have invented prognostic genetic signatures: lists of genes that can be used in scientific analyses to predict if a patient will survive or not. In this study, we joined together five different prognostic signatures, each of them related to a specific cancer type, to generate a unique pan-cancer prognostic signature, that contains 207 unique probesets related to 187 unique gene symbols, with one particular probeset present in two cancer type-specific signatures (203072_at related to the MYO1E gene). We applied our proposed pan-cancer signature with the Random Forests machine learning method to 57 microarray gene expression datasets of 12 different cancer types, and analyzed the results. We also compared the performance of our pan-cancer signature with the performances of two alternative prognostic signatures, and with the performances of each cancer type-specific signature on their corresponding cancer type-specific datasets. Our results confirmed the effectiveness of our prognostic pan-cancer signature. Moreover, we performed a pathway enrichment analysis, which indicated an association between the signature genes and a protein-protein interaction analysis, that highlighted PIK3R2 and FN1 as key genes having a fundamental relevance in our signature, suggesting an important role in pan-cancer prognosis for both of them.

© 2022. The Author(s).

In plain English

This computer-based study of gene activity in cancer datasets was not about asbestos, which appears only in the abstract as an example cause of cancer; no mesothelioma datasets were analysed. The authors merged five published survival gene lists (breast, lung, prostate and colon cancer, and neuroblastoma) into one 'pan-cancer' signature of 207 probesets linked to 187 genes. Using Random Forests, a machine learning method, they tested whether it separated survivors from deceased patients in 57 public datasets of 12 cancer types. Its main score (Matthews correlation coefficient) was sufficient on 19 of 57 datasets (33.33%). The authors reported adequate recognition of survivors on 58.18% of datasets and of deceased patients on 21.82%, but their own table shows 32 of 57 and 25 of 57 datasets at the stated cutoff. It beat the original cancer-specific lists on most colon, breast and lung datasets but lost on the single neuroblastoma dataset. It beat two other published pan-cancer signatures on 71.93% and 75.44% of datasets. The authors concluded that, even if not perfect, the signature's genes have a relevant role in cancer prognosis.

We wrote this plain-English summary from the full article with the help of AI, and a second AI model checked each sentence against the article. It is not the authors' wording; the abstract and the article are their own account.

What the study did and found

• Design: Computational (bioinformatics) study: a combined gene signature was applied with Random Forests machine learning to existing public gene expression datasets, followed by pathway and protein-interaction analyses
• Who was studied: 57 publicly available microarray gene expression datasets from Gene Expression Omnibus covering 12 cancer types, downloaded in April and May 2021. These were 17 breast, 13 lung, 10 colorectal, 5 lymphoma, 4 leukemia and 2 multiple myeloma datasets, plus 1 each of adrenocortical, bladder, neuroblastoma, ovarian, skin and stomach cancer. There were no mesothelioma datasets.
• Exposure: No asbestos or other environmental exposure was studied. The study examined the expression of a signature of 207 probesets (187 genes) merged from five cancer type-specific prognostic signatures.
• Main results:
  • The pan-cancer signature reached a sufficient MCC (+0.2 or higher) on 19 of 57 datasets (33.33%)
  • At least one performance score was sufficient on 55 of 57 datasets (all except dataMicke2011 and dataLeich2009)
  • The authors reported sufficient sensitivity (recognising survivors) on 58.18% of datasets and sufficient specificity (recognising deceased patients) on only 21.82%. Counting Table 2 against the stated cutoff of 0.6 or higher gives sensitivity on 32 of 57 datasets and specificity on 25 of 57, so the reported figures do not match the table.
  • Among the colon cancer datasets, MCC was above +0.2 on 6 of 10
  • The pan-cancer signature outperformed the colon cancer source signature (sigVanLaar2010) on 9 of 10 colon cancer datasets. It outperformed the lung cancer source signature (sigGyorffy2013) on 7 of 13 lung cancer datasets.
  • It was outperformed by the neuroblastoma source signature on the only neuroblastoma dataset
  • It outperformed two other published pan-cancer signatures: sigNagy2021 on 71.93% of datasets and sigYu2021 on 75.44%
  • The signature genes were enriched in 13 core pathways. In the protein-interaction network, PIK3R2 had the most physical interactions (5), followed by FN1 (4).
• The authors' conclusion: The authors concluded that, even if not perfect, the genes of their signature have a relevant role in pan-cancer prognosis and can serve as a starting point for future studies. They also concluded it may be less effective than cancer type-specific signatures for rare childhood cancers.
• Limitations the authors state:
  • The signature reached a sufficient MCC on only a minority of datasets
  • Only microarray data were used, not RNA-Seq data
  • TCGA and ArrayExpress data could not be used because their probesets were not compatible
  • No compatible prognostic prostate cancer dataset was found, so the prostate signature could not be tested on its own
  • The first pathway findings were biased by the unequal sizes of the five source signatures (the colon cancer signature made up 78.37%)

Topics

Medical terms in this study, explained in our Lexicon: Prognosis.

Related studies

• MicroRNA signature of malignant mesothelioma with potential diagnostic and prognostic implications — S. Busacca, S. Germano, et al. (2010)
• A gene signature linked to fibroblast differentiation for prognostic prediction of mesothelioma — J. Liu, Y. Lu, et al. (2024)
• The use of a next-generation sequencing-derived machine-learning risk-prediction model (OncoCast-MPM) for malignant pleural mesothelioma: a retrospective study — M.G. Zauderer, A. Martin, et al. (2021)
• Prediction of Radiation Pneumonitis With Machine Learning in Stage III Lung Cancer: A Pilot Study — M. Yakar, D. Etiz, et al. (2021)
• Predicting cancer using supervised machine learning: Mesothelioma — A. Choudhury (2021)
• Differential diagnosis of pleural mesothelioma using Logic Learning Machine — S. Parodi, R. Filiberti, et al. (2015)

Browse return to medical research.