A multi-class predictor based on a probabilistic model: application to gene expression profiling-based diagnosis of thyroid tumors

Our plain-English summary of A multi-class predictor based on a probabilistic model: application to gene expression profiling-based diagnosis of thyroid tumors, by N. Yukinawa, S. Oba, K. Kato, et al (BMC Genomics, 2006).

The study

• Authors: N. Yukinawa, S. Oba, K. Kato, K. Taniguchi, K. Iwao-Koizumi, Y. Tamaki, S. Noguchi, S. Ishii
• Journal: BMC Genomics
• Published: 2006;7:190
• Cited by: 11 articles, as counted by NIH iCite
• Read it: PubMed · Publisher · Free full text

Abstract

Background: Although microscopic diagnosis has been playing the decisive role in cancer diagnostics, there have been cases in which it does not satisfy the clinical need. Differential diagnosis of malignant and benign thyroid tissues is one such case, and supplementary diagnosis such as that by gene expression profile is expected.

Results: With four thyroid tissue types, i.e., papillary carcinoma, follicular carcinoma, follicular adenoma, and normal thyroid, we performed gene expression profiling with adaptor-tagged competitive PCR, a high-throughput RT-PCR technique. For differential diagnosis, we applied a novel multi-class predictor, introducing probabilistic outputs. Multi-class predictors were constructed using various combinations of binary classifiers. The learning set included 119 samples, and the predictors were evaluated by strict leave-one-out cross validation. Trials included classical combinations, i.e., one-to-one, one-to-the-rest, but the predictor using more combination exhibited the better prediction accuracy. This characteristic was consistent with other gene expression data sets. The performance of the selected predictor was then tested with an independent set consisting of 49 samples. The resulting test prediction accuracy was 85.7%.

Conclusion: Molecular diagnosis of thyroid tissues is feasible by gene expression profiling, and the current level is promising towards the automatic diagnostic tool to complement the present medical procedures. A multi-class predictor with an exhaustive combination of binary classifiers could achieve a higher prediction accuracy than those with classical combinations and other predictors such as multi-class SVM. The probabilistic outputs of the predictor offer more detailed information for each sample, which enables visualization of each sample in low-dimensional classification spaces. These new concepts should help to improve the multi-class classification including that of cancer tissues.

In plain English

This methodology article described a new computer method for sorting tissue samples into more than two diagnostic groups based on gene expression, the activity levels of genes. The authors measured 2516 genes in thyroid tissue from patients who had surgery between February 1998 and December 2002. The tissue types were papillary carcinoma, follicular carcinoma, follicular adenoma and normal thyroid. The method combined many two-way classifiers and gave a probability for each group. In a learning set of 119 samples, tested by leaving out one sample at a time, the best versions correctly classified 79.8% of samples, compared with 69.8% and 77.3% for two conventional combinations. In a separate test set of 49 samples, all three selected versions were 85.7% accurate. For comparison, the authors also ran the method on three other data sets. One of them included 11 mesothelioma samples among five cancer types. Asbestos exposure was not studied. The authors concluded that the approach showed promise for telling apart similar tumor types from the same tissue of origin, but said confirmatory experiments were needed before drawing a conclusion on its efficacy.

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: Methodology article: development and testing of a multi-class computer predictor using gene expression data from thyroid tissues, evaluated by leave-one-out cross-validation and an independent test set
• Who was studied: Thyroid tissue samples from patients who underwent hemithyroidectomy or total thyroidectomy, February 1998 to December 2002. Learning set: 41 follicular adenoma, 20 follicular carcinoma, 30 papillary carcinoma, 28 normal thyroid (119 samples). Test set: 17 follicular adenoma, 8 follicular carcinoma, 12 papillary carcinoma, 12 normal thyroid (49 samples).
• Exposure: No asbestos or other environmental exposure was studied. The data were expression levels of 2516 genes measured by adaptor-tagged competitive PCR (ATAC-PCR).
• Main results:
  • In the learning set (119 samples, leave-one-out cross-validation), the 1A and AA combinations were more accurate than the conventional one-to-one (11) and one-to-the-rest (1R) combinations. Table 1 gives 77.3% for 11 and 69.8% for 1R.
  • In the learning set, accuracy was highest for 1A-SIS and 1A-SHS (79.8%), followed by AA-SHS (79.0%).
  • In the learning set, the 1A-SHS predictor rarely misclassified papillary carcinoma.
  • In the independent test set of 49 samples, overall prediction accuracy was 85.7% for all three selected predictors (1A-SIS, 1A-SHS, AA-SHS).
  • Across the four data sets, the authors reported their method outperformed a multi-class support vector machine method and the shrunken centroid method.
  • One comparison data set (GCM) used five cancer types, including 11 mesothelioma samples. On that set, AA-SHS had the highest accuracy.
• The authors' conclusion: The authors concluded that the new approaches show promise for differentiating similar tumor types from the same tissue of origin.
• Limitations the authors state:
  • The authors stated that confirmatory experiments, including analysis of artificial data sets, are needed before concluding on the method's efficacy.
  • With the thyroid data, 1A was more accurate than AA, which the authors said suggests their probabilistic model is incomplete.
  • The number of classifiers grows exponentially as the number of classes increases.
  • The model assumes the choice of each binary classifier is independent, but the authors noted these classifiers may not be independent.

Topics

Medical terms in this study, explained in our Lexicon: Differential Diagnosis, Carcinoma.

Related studies

• Clinical validation of a 90-gene expression test for tumor tissue of origin diagnosis: a large-scale multicenter study of 1417 patients — W. Sun, W. Wu, et al. (2022)
• Utility of PAX8 mouse monoclonal antibody in the diagnosis of thyroid, thymic, pleural and lung tumours: a comparison with polyclonal PAX8 antibody — A. Toriyama, T. Mori, et al. (2014)
• Anaplastic thyroid carcinoma mimicking a malignant pleural mesothelioma: Clues for diagnosis — T. Capron, S. Giusiano, et al. (2019)
• Global gene expression profiling of human pleural mesotheliomas: identification of matrix metalloproteinase 14 (MMP-14) as potential tumour target — S. Crispi, R.A. Calogero, et al. (2009)
• Indications for distinct pathogenic mechanisms of asbestos and silica through gene expression profiling of the response of lung epithelial cells — T.N. Perkins, P.M. Peeters, et al. (2015)
• Gene expression profiling and gene copy-number changes in malignant mesothelioma cell lines — C. Zanazzi, R. Hersmus, et al. (2007)

Browse return to medical research.