engleski

Machine learning approach to the analysis of peptide immunomodulation in multiple sclerosis and optic neuritis.

Objectives: Peptide immunotherapy has been successfully applied as a therapeutic procedure for several immune-mediated diseases. Empirical observations showed that standard statistical approach is not an appropriate tool for the determination of prognostic parameters during peptide therapy. Therefore, we applied machine learning approach based on the C4.5 decision tree as a classifier. The method has been tested on the model of peptid-M (PENK_HUMAN 100-104 aa) vaccination in multiple sclerosis and optic neuritis. Methods: C4.5 decision tree has been tested on the model of peptid-M (Lupex) therapy in multiple sclerosis/optic neuritis. Results: Decision rules generated by the classifier extracted relationships between different parameters relevant for the prediction of beneficial peptide effects. The training set for the decision tree generator consisted of 38 tests observed before and one month after the peptide administration. Predictive parameters were EDSS, IFN, sIL-2R, sCD23 and peripheral blood cell populations CD20+23+, CD8+, CD8+beta2-M+, CD4+, CD4+b2-M+, CD4+25+ and CD3+16+56+. Conclusion: The accuracy of the procedure with respect to the therapy was 92- 100% for small samples. This model of non-linear prediction provided useful alternative to the standard statistical approach, enabled the extraction of few relevant parameters or their mutual relationships and ensured accurate prediction of the therapeutic procedure. The data were comparable to the clinical amelioration evaluated by the improvement of EDSS, VEP, colour vision, visual fields and MRI findings, 6 months and one year following the beginning of treatment.

peptide therapy; machine learning; decision tree; peptid-M; non-linear prediction; multiple sclerosis; optic neuritis

nije evidentirano