Artificial neural networks and decision tree model analysis of liver cancer proteomes

Application of artificial neural networks in detection and diagnosis of gastrointestinal and liver tumors

As a form of artificial intelligence, artificial neural networks (ANNs) have the advantages of adaptability, parallel processing capabilities, and non-linear processing. They have been widely used in the early detection and diagnosis of tumors. In this article, we introduce the development, working principle, and characteristics of ANNs and review the research progress on the application of ANNs in the detection and diagnosis of gastrointestinal and liver tumors.

Applying Data Mining to Investigate Cancer Risk in Patients with Pyogenic Liver Abscess

Pyogenic liver abscess is usually a complication of biliary tract disease. Taiwan features among the countries with the highest incidence of colorectal cancer (CRC) and hepatocellular carcinoma (HCC). Few studies have investigated whether patients with pyogenic liver abscess (PLA) have higher incidence rates of CRC and HCC. However, these findings have been inconclusive. The risks of CRC and HCC in patients with PLA and the factors contributing to cancer development were assessed in these patients. The clinical tests significantly associated with cancers in these patients with PLA were determined to assist in the early diagnosis of these cancers. Odds ratios (ORs) and 95% confidence intervals (CIs) were determined using binary logistic regression Cancer classification models were constructed using the decision tree algorithm C5.0 to compare the accuracy among different models with those risk factors of cancers and then determine the optimal model. Thereafter, the rules were summarized using the decisi8on tree model to assist in the diagnosis. The results indicated that CRC and HCC (OR, 3.751; 95% CI, 1.149–12.253) and CRC (OR, 6.838; 95% CI, 2.679–17.455) risks were higher in patients with PLA than those without PLA. The decision tree analysis demonstrated that the model with the PLA variable had the highest accuracy, and that classification could be conducted using fewer factors, indicating that PLA is critical in HCC and CRC. Two rules were determined for assisting in the diagnosis of CRC and HCC using the decision tree model.

Proteomics and phosphoproteomics in precision medicine: applications and challenges

Recent advances in proteomics allow the accurate measurement of abundances for thousands of proteins and phosphoproteins from multiple samples in parallel. Therefore, for the first time, we have the opportunity to measure the proteomic profiles of thousands of patient samples or disease model cell lines in a systematic way, to identify the precise underlying molecular mechanism and discover personalized biomarkers, networks and treatments. Here, we review examples of successful use of proteomics and phosphoproteomics data sets in as well as their integration other omics data sets with the aim of precision medicine. We will discuss the bioinformatics challenges posed by the generation, analysis and integration of such large data sets and present potential reasons why proteomics profiling and biomarkers are not currently widely used in the clinical setting. We will finally discuss ways to contribute to the better use of proteomics data in precision medicine and the clinical setting.

Improvement in Classification Performance Based on Target Vector Modification for All-Transfer Deep Learning

This paper proposes a target vector modification method for the all-transfer deep learning (ATDL) method. Deep neural networks (DNNs) have been used widely in many applications; however, the DNN has been known to be problematic when large amounts of training data are not available. Transfer learning can provide a solution to this problem. Previous methods regularize all layers, including the output layer, by estimating the relation vectors, which are then used instead of one-hot target vectors of the target domain. These vectors are estimated by averaging the target domain data of each target domain label in the output space. This method improves the classification performance, but it does not consider the relation between the relation vectors. From this point of view, we propose a relation vector modification based on constrained pairwise repulsive forces. High pairwise repulsive forces provide large distances between the relation vectors. In addition, the risk of divergence is mitigated by the constraint based on distributions of the output vectors of the target domain data. We apply our method to two simulation experiments and a disease classification using two-dimensional electrophoresis images. The experimental results show that reusing all layers through our estimation method is effective, especially for a significantly small number of the target domain data.

Analysis of significant protein abundance from multiple reaction-monitoring data

Background

Discovering reliable protein biomarkers is one of the most important issues in biomedical research. The ELISA is a traditional technique for accurate quantitation of well-known proteins. Recently, the multiple reaction-monitoring (MRM) mass spectrometry has been proposed for quantifying newly discovered protein and has become a popular alternative to ELISA. For the MRM data analysis, linear mixed modeling (LMM) has been used to analyze MRM data. MSstats is one of the most widely used tools for MRM data analysis that is based on the LMMs. However, LMMs often provide various significance results, depending on model specification. Sometimes it would be difficult to specify a correct LMM method for the analysis of MRM data. Here, we propose a new logistic regression-based method for Significance Analysis of Multiple Reaction Monitoring (LR-SAM).

Results

Through simulation studies, we demonstrate that LMM methods may not preserve type I error, thus yielding high false- positive errors, depending on how random effects are specified. Our simulation study also shows that the LR-SAM approach performs similarly well as LMM approaches, in most cases. However, LR-SAM performs better than the LMMs, particularly when the effects sizes of peptides from the same protein are heterogeneous. Our proposed method was applied to MRM data for identification of proteins associated with clinical responses of treatment of 115 hepatocellular carcinoma (HCC) patients with the tyrosine kinase inhibitor sorafenib. Of 124 candidate proteins, LMM approaches provided 6 results varying in significance, while LR-SAM, by contrast, yielded 18 significant results that were quite reproducibly consistent.

Conclusion

As exemplified by an application to HCC data set, LR-SAM more effectively identified proteins associated with clinical responses of treatment than LMM did.

Comparison of three data mining models for prediction of advanced schistosomiasis prognosis in the Hubei province

BACKGROUND

In order to better assist medical professionals, this study aimed to develop and compare the performance of three models-a multivariate logistic regression (LR) model, an artificial neural network (ANN) model, and a decision tree (DT) model-to predict the prognosis of patients with advanced schistosomiasis residing in the Hubei province.

METHODOLOGY/PRINCIPAL FINDINGS

Schistosomiasis surveillance data were collected from a previous study based on a Hubei population sample including 4136 advanced schistosomiasis cases. The predictive models use LR, ANN, and DT methods. From each of the three groups, 70% of the cases (2896 cases) were used as training data for the predictive models. The remaining 30% of the cases (1240 cases) were used as validation groups for performance comparisons between the three models. Prediction performance was evaluated using area under the receiver operating characteristic curve (AUC), sensitivity, specificity, and accuracy. Univariate analysis indicated that 16 risk factors were significantly associated with a patient's outcome of prognosis. In the training group, the mean AUC was 0.8276 for LR, 0.9267 for ANN, and 0.8229 for DT. In the validation group, the mean AUC was 0.8349 for LR, 0.8318 for ANN, and 0.8148 for DT. The three models yielded similar results in terms of accuracy, sensitivity, and specificity.

CONCLUSIONS/SIGNIFICANCE

Predictive models for advanced schistosomiasis prognosis, respectively using LR, ANN and DT models were proved to be effective approaches based on our dataset. The ANN model outperformed the LR and DT models in terms of AUC.

Predicting metachronous liver metastasis from colorectal cancer using serum proteomic fingerprinting

BACKGROUND

There are currently no accurate predictive markers of metachronous liver metastasis (MLM) from colorectal cancer.

METHODS

Magnetic bead-based fractionation coupled with mass spectrometry analysis was used to compare serum samples from 64 patients with MLM and 64 without recurrence or metastasis for at least 3 y after radical colorectal surgery (NM). A total of 40 MLM and 40 NM serum samples were randomly selected to build a decision tree, and the remainder were tested as blinded samples. Selected peptides were identified.

RESULTS

The patients in the two groups were matched for gender, age, tumor location, TNM staging, and histologic differentiation grade. Preoperative serum carcinoembryonic antigen retained no independent power to predict MLM. The decision tree model with eight proteomic features (m/z 3315, 6637, 1207, 1466, 4167, 4210, 2660, and 4186) correctly classified 33 of 40 NM sera (82.5%) and 32 of 40 MLM sera (80%) in the training set and 19 of 24 NM sera (79.2%) and 17 of 24 MLM sera (70.8%) in the test set. The peptides were identified as fragments of alpha-fetoprotein, complement C4-A, fibrinogen alpha, eukaryotic peptide chain release factor GTP-binding subunit ERF3B, and angiotensinogen.

CONCLUSIONS

In patients matched for gender, age, tumor location, TNM staging, and histologic differentiation grade, preoperative carcinoembryonic antigen retained no independent power to predict MLM. The decision tree model of eight proteomic features demonstrated promising value for predicting MLM in patients who underwent radical resection of colorectal cancer.

Mortality predicted accuracy for hepatocellular carcinoma patients with hepatic resection using artificial neural network

The aim of this present study is firstly to compare significant predictors of mortality for hepatocellular carcinoma (HCC) patients undergoing resection between artificial neural network (ANN) and logistic regression (LR) models and secondly to evaluate the predictive accuracy of ANN and LR in different survival year estimation models. We constructed a prognostic model for 434 patients with 21 potential input variables by Cox regression model. Model performance was measured by numbers of significant predictors and predictive accuracy. The results indicated that ANN had double to triple numbers of significant predictors at 1-, 3-, and 5-year survival models as compared with LR models. Scores of accuracy, sensitivity, specificity, and area under the receiver operating characteristic curve (AUROC) of 1-, 3-, and 5-year survival estimation models using ANN were superior to those of LR in all the training sets and most of the validation sets. The study demonstrated that ANN not only had a great number of predictors of mortality variables but also provided accurate prediction, as compared with conventional methods. It is suggested that physicians consider using data mining methods as supplemental tools for clinical decision-making and prognostic evaluation.

Disease-free survival after hepatic resection in hepatocellular carcinoma patients: a prediction approach using artificial neural network

Background

A database for hepatocellular carcinoma (HCC) patients who had received hepatic resection was used to develop prediction models for 1-, 3- and 5-year disease-free survival based on a set of clinical parameters for this patient group.

Methods

The three prediction models included an artificial neural network (ANN) model, a logistic regression (LR) model, and a decision tree (DT) model. Data for 427, 354 and 297 HCC patients with histories of 1-, 3- and 5-year disease-free survival after hepatic resection, respectively, were extracted from the HCC patient database. From each of the three groups, 80% of the cases (342, 283 and 238 cases of 1-, 3- and 5-year disease-free survival, respectively) were selected to provide training data for the prediction models. The remaining 20% of cases in each group (85, 71 and 59 cases in the three respective groups) were assigned to validation groups for performance comparisons of the three models. Area under receiver operating characteristics curve (AUROC) was used as the performance index for evaluating the three models.

Conclusions

The ANN model outperformed the LR and DT models in terms of prediction accuracy. This study demonstrated the feasibility of using ANNs in medical decision support systems for predicting disease-free survival based on clinical databases in HCC patients who have received hepatic resection.

Algorithm to determine the outcome of patients with acute liver failure: a data-mining analysis using decision trees

BACKGROUND

We established algorithms to predict the prognosis of acute liver failure (ALF) patients through a data-mining analysis, in order to improve the indication criteria for liver transplantation.

METHODS

The subjects were 1,022 ALF patients seen between 1998 and 2007 and enrolled in a nationwide survey. Patients older than 65 years, and those who had undergone liver transplantation and received blood products before the onset of hepatic encephalopathy were excluded. Two data sets were used: patients seen between 1998 and 2003 (n=698), whose data were used for the formation of the algorithm, and those seen between 2004 and 2007 (n=324), whose data were used for the validation of the algorithm. Data on a total of 73 items, at the onset of encephalopathy and 5 days later, were collected from 371 of the 698 patients seen between 1998 and 2003, and their outcome was analyzed to establish decision trees. The obtained algorithm was validated using the data of 160 of the 324 patients seen between 2004 and 2007.

RESULTS

The outcome of the patients at the onset of encephalopathy was predicted through 5 items, and the patients were classified into 6 categories with mortality rates between 23% and89%. When the prognosis of the patients in the categories with mortality rates greater than 50% was predicted as "death", the accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) of the algorithm were 79, 78, 81, 83, and 75%, respectively. Similar high values were obtained when the algorithm was employed in the patients for validation. The outcome of the patients 5 days after the onset of encephalopathy was predicted through 7 items, and a similar high accuracy was found for both sets of patients.

CONCLUSIONS

Novel algorithms for predicting the outcome of ALF patients may be useful to determine the indication for liver transplantation.

Discovery of lamin B1 and vimentin as circulating biomarkers for early hepatocellular carcinoma

The recent advancements in proteomic technologies have reconstituted our research strategies over different type of liver diseases including hepatocellular carcinoma (HCC). Combined analyses on HCC proteome and clinicopathological data of patients have allowed identification of many promising biomarkers that can be further developed into noninvasive diagnostic assays for cancer surveillance. Capitalizing our established proteomic platform primarily based on two-dimensional polyacrylamide gel electrophoresis (2DE) and MALDI-TOF/TOF mass spectrometry, our groups have identified lamin B1 (LMNB1) and vimentin (VIM) as promising biomarkers for detection of early HCC. Protein levels of both biomarkers were significantly elevated in cancerous tissues when compared to the controls in disease-free and cirrhotic liver subjects. Further investigation of the circulating LMNB1 mRNA level in patients' blood samples by standard PCR showed 76% sensitivity and 82% specificity for detection of early HCC. In parallel, an ELISA assay for measuring circulating vimentin level in patients' serum samples could detect small HCC at 40.91% sensitivity and 87.5% specificity. The candidate biomarkers were evaluated with the diagnostic performance of α-fetoprotein (AFP) for HCC. In this article, we address the current protocols for HCC biomarker discovery, ranging from clinical sample preparation, 2DE proteomic profiling and informatics analysis, and assay development and clinical validation study. Focus is emphasized on the methods for sample preservation and low-abundance protein enrichment.

Case-based reasoning support for liver disease diagnosis

OBJECTIVES

In Taiwan, as well as in the other countries around the world, liver disease has reigned over the list of leading causes of mortality, and its resistance to early detection renders the disease even more threatening. It is therefore crucial to develop an auxiliary system for diagnosing liver disease so as to enhance the efficiency of medical diagnosis and to expedite the delivery of proper medical treatment.

METHODS

The study accordingly integrated the case-based reasoning (CBR) model into several common classification methods of data mining techniques, including back-propagation neural network (BPN), classification and regression tree, logistic regression, and discriminatory analysis, in an attempt to develop a more efficient model for early diagnosis of liver disease and to enhance classification accuracy. To minimize possible bias, this study used a ten-fold cross-validation to select a best model for more precise diagnosis results and to reduce problems caused by false diagnosis.

RESULTS

Through a comparison of five single models, BPN and CBR emerged to be the top two methods in terms of overall performance. For enhancing diagnosis performance, CBR was integrated with other methods, and the results indicated that the accuracy and sensitivity of each CBR-added hybrid model were higher than those of each single model. Of all the CBR-added hybrid models, the BPN-CBR method took the lead in terms of diagnosis capacity with an accuracy rate of 95%, a sensitivity of 98%, and a specificity of 94%.

CONCLUSIONS

After comparing the five single and hybrid models, the study found BPN-CBR the best model capable of helping physicians to determine the existence of liver disease, achieve an accurate diagnosis, diminish the possibility of a false diagnosis being given to sick people, and avoid the delay of clinical treatment.

Sulforaphane induces apoptosis in human hepatic cancer cells through inhibition of 6-phosphofructo-2-kinase/fructose-2,6-biphosphatase4, mediated by hypoxia inducible factor-1-dependent pathway

The anti-cancer activity of sulforaphane (SFN) has recently been investigated in several cancer cell lines, including human hepatic cancers. However, the mechanism of SFN-induced cell death in human hepatic cancer cells is still not well understood. The aim of the present work is to explore the possible mechanisms of SFN-induced apoptosis in hepatocellular carcinoma cells using proteomic analysis. A two-dimensional electrophoresis (2-DE)-based-proteomic analysis was employed for identification of possible target-related proteins of SFN-induced apoptosis. Among eleven proteins identified as regulated, we focused on the down-regulation of 6-phosphofructo-2-kinase/fructose-2,6-biphosphatase4 (PFKFB4) protein, which has been known as a key modulator of glycolysis. We also showed that SFN down-regulated the expression of the transcriptional factor, hypoxia inducible factor-1α (HIF-1α), which strongly regulates PFKFB4 expression. In order to obtain a broad understanding of the correlation of HIF-1α and SFN, we observed the inhibition of the activity of mitogen-activated protein kinases, regulators of HIF-1α activity. Our findings suggest that SFN is a potent inducer of apoptosis in hepatocellular carcinoma cells via PFKFB4-inhibition pathways. HIF-1 pathway inhibition may be mediated by the inhibition of mitogen-activated protein kinases.

Proteomics of hepatocellular carcinoma in Chinese patients

Hepatocellular carcinoma (HCC) is a malignant tumor of liver that causes approximately half a million deaths each year, of which over half of the cases are diagnosed in China. Because of its asymptomatic nature, HCC is usually diagnosed at late and advanced stages, for which there are no effective therapies. Thus, biomarkers for early detection and molecular targets for treating HCC are urgently needed. With the advent of high-throughput omics technologies, we have begun to mine the genomics and proteomics information of HCC, and most importantly, these data can be integrated with clinical annotations of the patients. Such new horizons of integrated profiling informatics have allowed us to search for and better identify clinically useful biomarkers and therapeutic targets for cancers including HCC. Capitalizing the large clinical samples cohort (over 100 pairs of tumor and matched adjacent nontumor tissues of HCC), we herein discuss the use of proteomics approach to identify biomarkers that are potentially useful for (1) discrimination of tumorous from nonmalignant tissues, (2) detection of small-sized and early stage of HCC, and (3) prediction of early disease relapse after hepatectomy.

An Analysis of the Correlation between Alopecia and Chief Complaints

Objectives

In this study, we measured the extent of ten levels of classified symptoms by 300 (male and female) patients visiting the hair loss clinics of "S" hospitals in Gangbuk and Gangnam between January 2009 and June 2011 by analyzing the patients' chief complaints.

Methods

The method of measurement was based on a symptom questionnaire possessing 51 categories. Through the statistical analysis of data mining techniques, decision trees, and logistic regression, we derived a logistic regression model and decision tree model that improved both the response rate and significant hair loss-related characteristics of the questionnaire.

Results

The results of this study indicate that dry hair, seborrheic scalps and skin, tobacco and/or coffee addiction, anxiety, nausea, indigestion, and facial flushing correlate to hair loss.

Conclusions

We anticipate that the subjective symptoms of hair loss can provide a foundation for preventing secondary diseases and provide clinical data information during the period of treatment. This can contribute to the improvement of patient satisfaction after customized treatment.

Neural networks

Neural networks are a class of intelligent learning machines establishing the relationships between descriptors of real-world objects. As optimisation tools they are also a class of computational algorithms implemented using statistical/numerical techniques for parameter estimate, model selection, and generalisation enhancement. In bioinformatics applications, neural networks have played an important role for classification, function approximation, knowledge discovery, and data visualisation. This chapter will focus on supervised neural networks and discuss their applications to bioinformatics.

Enhanced detection of early hepatocellular carcinoma by serum SELDI-TOF proteomic signature combined with alpha-fetoprotein marker

Background

Biomarkers for accurate diagnosis of early hepatocellular carcinoma (HCC) are limited in number and clinical validation. We applied SELDI-TOF-MS ProteinChip technology to identify serum profile for distinguishing HCC and liver cirrhosis (LC) and to compare the accuracy of SELDI-TOF-MS profile and alpha-fetoprotein (AFP) level in HCC diagnosis.

Patients and Methods

Serum samples were obtained from 120 HCC and 120 LC patients for biomarker discovery and validation studies. ProteinChip technology was employed for generating SELDI-TOF proteomic features and analyzing serum proteins/peptides.

Results

A diagnostic model was established by CART algorithm, which is based on 5 proteomic peaks with m/z values at 3324, 3994, 4665, 4795, and 5152. In the training set, the CART algorithm could differentiate HCC from LC subjects with a sensitivity and specificity of 98% and 95%, respectively. The results were assessed in blind validation using separate cohorts of 60 HCC and 60 LC patients, with an accuracy of 83% for HCC and 92% for LC patients. The diagnostic odd ratio (DOR) indicated that SELDI-TOF proteomic signature could achieve better diagnostic performance than serum AFP level at a cutoff of 20 ng/mL (AFP₂₀) (92.72 vs 9.11), particularly superior for early-stage HCC (87% vs 54%). Importantly, a combined use of both tests could enhance the detection of HCC (sensitivity, 95%; specificity, 98%; DOR, 931).

Conclusion

Serum SELDI-TOF proteomic signature, alone or in combination with AFP marker, promises to be a good tool for early diagnosis and/screening of HCC in at-risk population with liver cirrhosis.

Proteomic expression signature distinguishes cancerous and nonmalignant tissues in hepatocellular carcinoma

Hepatocellular carcinoma (HCC) is an aggressive liver cancer but clinically validated biomarkers that can predict natural history of malignant progression are lacking. The present study explored the proteome-wide patterns of HCC to identify biomarker signature that could distinguish cancerous and nonmalignant liver tissues. A retrospective cohort of 80 HBV-associated HCC was included and both the tumor and adjacent nontumor tissues were subjected to proteome-wide expression profiling by 2-DE method. The subjects were randomly divided into the training (n = 55) and validation (n = 25) subsets, and the data analyzed by classification-and-regression tree algorithm. Protein markers were characterized by MALDI-ToF/MS and confirmed by immunohistochemistry, Western blotting and qPCR assays. Proteomic expression signature composed of six biomarkers (haptoglobin, cytochrome b5, progesterone receptor membrane component 1, heat shock 27 kDa protein 1, lysosomal proteinase cathepsin B, keratin I) was developed as a classifier model for predicting HCC. We further evaluated the model using both leave-one-out procedure and independent validation, and the overall sensitivity and specificity for HCC both are 92.5%, respectively. Clinical correlation analysis revealed that these biomarkers were significantly associated with serum AFP, total protein levels and the Ishak's score. The described model using biomarker signatures could accurately distinguish HCC from nonmalignant tissues, which may also provide hints on how normal hepatocytes are transformed to malignant state during tumor progression.

An intelligent model for liver disease diagnosis

OBJECTIVES

Liver disease, the most common disease in Taiwan, is not easily discovered in its initial stage; early diagnosis of this leading cause of mortality is therefore highly important. The design of an effective diagnosis model is therefore an important issue in liver disease treatment. This study accordingly employs classification and regression tree (CART) and case-based reasoning (CBR) techniques to structure an intelligent diagnosis model aiming to provide a comprehensive analytic framework to raise the accuracy of liver disease diagnosis.

METHODS

Based on the advice and assistance of doctors and medical specialists of liver conditions, 510 outpatient visitors using ICD-9 (International Classification of Diseases, 9th Revision) codes at a medical center in Taiwan from 2005 to 2006 were selected as the cases in the data set for liver disease diagnosis. Data on 340 patients was utilized for the development of the model and on 170 patients utilized to perform comparative analysis of the models. This paper accordingly suggests an intelligent model for the diagnosis of liver diseases which integrates CART and CBR. The major steps in applying the model include: (1) adopting CART to diagnose whether a patient suffers from liver disease; (2) for patients diagnosed with liver disease in the first step, employing CBR to diagnose the types of liver diseases.

RESULTS

In the first phase, CART is used to extract rules from health examination data to show whether the patient suffers from liver disease. The results indicate that the CART rate of accuracy is 92.94%. In the second phase, CBR is developed to diagnose the type of liver disease, and the new case triggers the CBR system to retrieve the most similar case from the case base in order to support the treatment of liver disease. The new case is supported by a similarity ratio, and the CBR diagnostic accuracy rate is 90.00%. Actual implementation shows that the intelligent diagnosis model is capable of integrating CART and CBR techniques to examine liver diseases with considerable accuracy. The model can be used as a supporting system in making decisions regarding liver disease diagnosis and treatment. The rules extracted from CART are helpful to physicians in diagnosing liver diseases. CBR can retrieve the most similar case from the case base in order to solve a new liver disease problem and can be of great assistance to physicians in identifying the type of liver disease, reducing diagnostic errors and improving the quality and effectiveness of medical treatment.

An introduction to artificial neural networks in bioinformatics--application to complex microarray and mass spectrometry datasets in cancer studies

Applications of genomic and proteomic technologies have seen a major increase, resulting in an explosion in the amount of highly dimensional and complex data being generated. Subsequently this has increased the effort by the bioinformatics community to develop novel computational approaches that allow for meaningful information to be extracted. This information must be of biological relevance and thus correlate to disease phenotypes of interest. Artificial neural networks are a form of machine learning from the field of artificial intelligence with proven pattern recognition capabilities and have been utilized in many areas of bioinformatics. This is due to their ability to cope with highly dimensional complex datasets such as those developed by protein mass spectrometry and DNA microarray experiments. As such, neural networks have been applied to problems such as disease classification and identification of biomarkers. This review introduces and describes the concepts related to neural networks, the advantages and caveats to their use, examples of their applications in mass spectrometry and microarray research (with a particular focus on cancer studies), and illustrations from recent literature showing where neural networks have performed well in comparison to other machine learning methods. This should form the necessary background knowledge and information enabling researchers with an interest in these methodologies, but not necessarily from a machine learning background, to apply the concepts to their own datasets, thus maximizing the information gain from these complex biological systems.

Examining the significance of fingerprint-based classifiers

Background

Experimental examinations of biofluids to measure concentrations of proteins or their fragments or metabolites are being explored as a means of early disease detection, distinguishing diseases with similar symptoms, and drug treatment efficacy. Many studies have produced classifiers with a high sensitivity and specificity, and it has been argued that accurate results necessarily imply some underlying biology-based features in the classifier. The simplest test of this conjecture is to examine datasets designed to contain no information with classifiers used in many published studies.

Results

The classification accuracy of two fingerprint-based classifiers, a decision tree (DT) algorithm and a medoid classification algorithm (MCA), are examined. These methods are used to examine 30 artificial datasets that contain random concentration levels for 300 biomolecules. Each dataset contains between 30 and 300 Cases and Controls, and since the 300 observed concentrations are randomly generated, these datasets are constructed to contain no biological information. A modest search of decision trees containing at most seven decision nodes finds a large number of unique decision trees with an average sensitivity and specificity above 85% for datasets containing 60 Cases and 60 Controls or less, and for datasets with 90 Cases and 90 Controls many DTs have an average sensitivity and specificity above 80%. For even the largest dataset (300 Cases and 300 Controls) the MCA procedure finds several unique classifiers that have an average sensitivity and specificity above 88% using only six or seven features.

Conclusion

While it has been argued that accurate classification results must imply some biological basis for the separation of Cases from Controls, our results show that this is not necessarily true. The DT and MCA classifiers are sufficiently flexible and can produce good results from datasets that are specifically constructed to contain no information. This means that a chance fitting to the data is possible. All datasets used in this investigation are available on the web.

This work is funded by NCI Contract N01-CO-12400.

Diagnosis of gastric cancer using near-infrared Raman spectroscopy and classification and regression tree techniques

The purpose of this study is to apply near-infrared (NIR) Raman spectroscopy and classification and regression tree (CART) techniques for identifying molecular changes of tissue associated with cancer transformation. A rapid-acquisition NIR Raman system is utilized for tissue Raman spectroscopic measurements at 785-nm excitation. 73 gastric tissue samples (55 normal, 18 cancer) from 53 patients are measured. The CART technique is introduced to develop effective diagnostic algorithms for classification of Raman spectra of different gastric tissues. 80% of the Raman dataset are randomly selected for spectral learning, while 20% of the dataset are reserved for validation. High-quality Raman spectra in the range of 800 to 1800 cm(-1) are acquired from gastric tissue within 5 s. The diagnostic sensitivity and specificity of the learning dataset are 90.2 and 95.7%; and the predictive sensitivity and specificity of the independent validation dataset are 88.9 and 92.9%, respectively, for separating cancer from normal. The tissue Raman peaks at 875 and 1745 cm(-1) are found to be two of the most significant features to discriminate gastric cancer from normal tissue. NIR Raman spectroscopy in conjunction with the CART technique has the potential to provide an effective and accurate diagnostic means for cancer detection in the gastric system.