Diagnostic classification of schizophrenia by neural network analysis of blood-based gene expression signatures

• Published in: Schizophrenia research Vol. 119; no. 1; pp. 210 - 218
• Main Authors: Takahashi, Makoto, Hayashi, Hiroshi, Watanabe, Yuichiro, Sawamura, Kazushi, Fukui, Naoki, Watanabe, Junzo, Kitajima, Tsuyoshi, Yamanouchi, Yoshio, Iwata, Nakao, Mizukami, Katsuyoshi, Hori, Takafumi, Shimoda, Kazutaka, Ujike, Hiroshi, Ozaki, Norio, Iijima, Kentarou, Takemura, Kazuo, Aoshima, Hideyuki, Someya, Toshiyuki
• Format: Journal Article
• Language: English
• Published: Amsterdam Elsevier B.V 01.06.2010; Elsevier
• Subjects: Adult; Adult and adolescent clinical studies; Artificial neural network; Bioinformatics; Biological and medical sciences; Biomarker; Brain - metabolism; cDNA microarray; Female; Gene Expression Profiling; Genetic Markers - genetics; Humans; Male; Medical sciences; Middle Aged; Neural Networks (Computer); Oligonucleotide Array Sequence Analysis; Psychiatry; Psychology. Psychoanalysis. Psychiatry; Psychopathology. Psychiatry; Psychoses; Psychotic Disorders - classification; Psychotic Disorders - diagnosis; Psychotic Disorders - genetics; Schizophrenia; Schizophrenia - classification; Schizophrenia - diagnosis; Schizophrenia - genetics; Sensitivity and Specificity; Young Adult; Schizophrenia; cDNA microarray; Biomarker; Artificial neural network; Bioinformatics; Human; Biological marker; Neural network; Gene expression; Genetic determinism; Psychosis; Complementary DNA; Classification; Genetics; Diagnosis
• ISSN: 0920-9964; 1573-2509
• DOI: 10.1016/j.schres.2009.12.024

Abstract Gene expression profiling with microarray technology suggests that peripheral blood cells might be a surrogate for postmortem brain tissue in studies of schizophrenia. The development of an accessible peripheral biomarker would substantially help in the diagnosis of this disease. We used a bioinformatics approach to examine whether the gene expression signature in whole blood contains enough information to make a specific diagnosis of schizophrenia. Unpaired t -tests of gene expression datasets from 52 antipsychotics-free schizophrenia patients and 49 normal controls identified 792 differentially expressed probes. Functional profiling with DAVID revealed that eleven of these genes were previously reported to be associated with schizophrenia, and 73 of them were expressed in the brain tissue. We analyzed the datasets with one of the supervised classifiers, artificial neural networks (ANNs). The samples were subdivided into training and testing sets. Quality filtering and stepwise forward selection identified 14 probes as predictors of the diagnosis. ANNs were then trained with the selected probes as the input and the training set for known diagnosis as the output. The constructed model achieved 91.2% diagnostic accuracy in the training set and 87.9% accuracy in the hold-out testing set. On the other hand, hierarchical clustering, a standard but unsupervised classifier, failed to separate patients and controls. These results suggest analysis of a blood-based gene expression signature with the supervised classifier, ANNs, might be a diagnostic tool for schizophrenia.