Diagnostic classification of schizophrenia by neural network analysis of blood-based gene expression signatures
Introduction
Genome-wide gene expression analysis by DNA microarray and bioinformatics procedures has been conducted to elucidate common gene pathways that underpin the biological mechanisms of schizophrenia (Aston et al., 2004, Hakak et al., 2001, Hemby et al., 2002, Iwamoto et al., 2005, Iwamoto et al., 2004, Mirnics et al., 2000, Mirnics et al., 2001, Sugai et al., 2004, Tkachev et al., 2003). However, the clinical use of microarray technology is not so widespread in schizophrenia research as compared with cancer research (Rhodes et al., 2004), due to the difficulty in interpreting results obtained from postmortem brain tissue that are complicated by agonal factors, anatomical inconsistency, and cellular heterogeneity of the cortical and subcortical regions. Postmortem brain studies use less accessible materials and therefore are limited by small sample size and repeated use of the same cohorts (Iwamoto and Kato, 2006). As a more accessible tissue, several researchers have undertaken expression profiling of peripheral blood cells (Sullivan et al., 2006, Tsuang et al., 2005, Vawter et al., 2004, Zvara et al., 2005). On a transcriptional expression level, peripheral blood cells were reported to share significant similarities with tissues from multiple brain regions (Sullivan et al., 2006). Interestingly, Tsuang et al., 2005, Middleton et al., 2005 have shown that a set of genes extracted from gene expression signature of isolated peripheral blood cells can discriminate between schizophrenia and control groups.
These studies suggest that analysis of high dimensional data is useful to generate a biomarker of schizophrenia since it can combine data from several molecules, each of which shows small difference but is not exclusively associated with this disease (Schwarz and Bahn, 2008). In cancer research, classification by gene expression signature is widely used to predict tumor classes, drug responses, and prognosis of individual subjects (Khan et al., 2001, Lin et al., 2007, O'Neill and Song, 2003). Development of such classifier will greatly help our diagnosis of schizophrenia that is solely dependent on clinical symptoms so far. There are two approaches in classification: supervised and unsupervised methods. In contrast to unsupervised clustering, supervised classifiers learn a function from training data that consist of pairs of input objects (e.g., gene expression signatures) and desired outputs (e.g., diagnoses) (De Bruyne et al., 2007). The artificial neural network (ANN) is one of those classifiers that works very well, at identifying patterns or trends in a large amount of data with little theory.
Purpose of the present study is to examine whether microarray date obtained from whole blood cells contain enough information to classify schizophrenia. We present here that ANN model can correctly predict the diagnosis with sufficient accuracy.
Section snippets
Subjects
Samples from 52 patients with schizophrenia and 49 normal controls were analyzed. Patients with schizophrenia or schizophreniform disorder were recruited from outpatients or inpatients of psychiatry unit at 6 centers across Japan. Those who were antipsychotics-free and had no comorbidity were included in the study. Control subjects were recruited from hospital staff and student volunteers who showed no evidence of present or past mental illness. All subjects were evaluated using the Structured
Characteristics and annotation of a blood-based gene expression signature
Among 19,121 quality filtered probes, 792 probes were identified as differentially expressed; 256 probes were down-regulated and 536 probes were up-regulated in patients with schizophrenia compared with controls. Since the gender ratio was different between the two groups, our filtering process excluded probes with significant gender differences. Down-regulated and up-regulated probes were separately profiled by DAVID. Among 256 down-regulated probes, 167 genes were annotated. Top-ranked Gene
Discussion
Using supervised method, ANN, we found that schizophrenia can be classified by blood-based gene expression signature. Accuracy of the ANN model was 91.2%, which was not so high as similar reports in cancer research (Khan et al., 2001, Lin et al., 2007, O'Neill and Song, 2003). However, in contrast to the use of tumor samples in those studies, we are not able to obtain tissues from the brain. Considering the use of peripheral whole blood as a material, we feel that the performance of our ANN
Role of funding source
This work was supported by the Contract Development Program of Japan Science and Technology Agency; the agency had no further role in study design; in the collection, analysis and interpretation of data; in the writing of the report; and in the decision to submit the paper for publication.
Contributors
Someya, Aoshima and Watanabe Y designed the study and wrote the protocol. Takahashi, Sawamura, Fukui, Watanabe J, Kitajima, Yamanouchi, Iwata, Mizukami, Hori, Shimoda, Ujike and Ozaki recruited subjects. Iijima and Takemura performed gene expression profiling, and Hayashi undertook the statistical analysis. Takahashi wrote the first draft of the manuscript. All authors contributed to and have approved the final manuscript.
Conflict of interest
Hayashi, Iijima, Takamura and Aoshima work for R&D Department of SRL Inc. All other authors declare that they have no conflicts of interest.
Acknowledgement
We thank Dr. Hiroyuki Nawa for the helpful discussions.
References (39)
- et al.
Microarray analysis of postmortem temporal cortex from patients with schizophrenia
J. Neurosci. Res.
(2004) - et al.
Natural variation in human gene expression assessed in lymphoblastoid cells
Nat. Genet.
(2003) - et al.
Genetic and physiological data implicating the new human gene G72 and the gene for d-amino acid oxidase in schizophrenia
Proc. Natl. Acad. Sci. U. S. A.
(2002) - et al.
Methods for microarray data analysis
Methods Mol. Biol.
(2007) - et al.
Comparison of different isolation techniques prior gene expression profiling of blood derived cells: impact on physiological responses, on overall expression and the role of different cell types
Pharmacogenomics J.
(2004) - et al.
Variation in gene expression profiles of peripheral blood mononuclear cells from healthy volunteers
Physiol. Genomics
(2005) - et al.
Genome-wide expression analysis reveals dysregulation of myelination-related genes in chronic schizophrenia
Proc. Natl. Acad. Sci. U. S. A.
(2001) - et al.
Decreased serum levels of d-serine in patients with schizophrenia: evidence in support of the N-methyl-d-aspartate receptor hypofunction hypothesis of schizophrenia
Arch. Gen. Psychiatry
(2003) - et al.
Gene expression profile for schizophrenia: discrete neuron transcription patterns in the entorhinal cortex
Arch. Gen. Psychiatry
(2002) - et al.
DAVID Bioinformatics Resources: expanded annotation database and novel algorithms to better extract biology from large gene lists
Nucleic Acids Res.
(2007)
Gene expression profiling in schizophrenia and related mental disorders
Neuroscientist
Molecular characterization of bipolar disorder by comparing gene expression profiles of postmortem brains of major mental disorders
Mol. Psychiatry
Altered expression of mitochondria-related genes in postmortem brains of patients with bipolar disorder or schizophrenia, as revealed by large-scale DNA microarray analysis
Hum. Mol. Genet.
Improved gene selection for classification of microarrays
Pac. Symp. Biocomput.
Classification and diagnostic prediction of cancers using gene expression profiling and artificial neural networks
Nat. Med.
Identification of gene transcript signatures predictive for estrogen receptor and lymph node status using a stepwise forward selection artificial neural network modelling approach
Artif. Intell. Med.
Improvement of bias and generalizability for computer-aided diagnostic schemes
Comput. Med. Imaging Graph.
Discovery of significant rules for classifying cancer diagnosis data
Bioinformatics
Multiple gene expression classifiers from different array platforms predict poor prognosis of colorectal cancer
Clin. Cancer Res.
Cited by (62)
Transcriptomics and machine learning to advance schizophrenia genetics: A case-control study using post-mortem brain data
2022, Computer Methods and Programs in BiomedicineDifferential blood-based biomarkers of psychopathological dimensions of schizophrenia
2016, Revista de Psiquiatria y Salud MentalAltered neural signaling and immune pathways in peripheral blood mononuclear cells of schizophrenia patients with cognitive impairment: A transcriptome analysis
2016, Brain, Behavior, and ImmunityCitation Excerpt :In simple gene level analysis, the ‘cognitive deficit’ (CD) subtype showed 130 DEGs, 106 of which were also found to be differentially expressed in the SZ group as a whole; however, the ‘cognitively spared’ (CS) subgroup displayed very little difference to the control group (HC) with only 4 DEGs surviving multiple correction (ZNF827, HINT2, LOC653103, YWHAB). Functional analysis of the DEGs using multiple methods detected several important biological themes including altered metabolism, immune response, and cell signaling in both CD and SZ, consistent with previous microarray studies (Kurian et al., 2011; Kuzman et al., 2009; Martin et al., 2009; Middleton et al., 2005b; Takahashi et al., 2010; Tsuang et al., 2005; Vawter et al., 2004b). In addition, using GAGE we detected 27 significantly altered pathways (19 pathways up-regulated and 8 down-regulated) in the combined SZ group and a further 6 pathways up-regulated in the CS group and 5 altered pathways (4 down-regulated and 1 up-regulated) in the CD group (Table 2).
A primer on the use of machine learning to distil knowledge from data in biological psychiatry
2024, Molecular Psychiatry
- 1
These authors contributed equally to this work.