Blind Test Set Research Articles

Preliminary study of serum proteome biomarkers of hypopharyngeal squamous cell carcinoma

To screen the serum proteome biomarkers of hypopharyngeal squamous cell carcinoma (HSCC) and to establish a predictive model for early detection of HSCC. Serum samples were collected from 48 HSCC patients before surgery and 52 age and sex-matched individuals without cancer used as controls. The samples were divided into 2 sets: training set (including 36 HSCC patients and 36 controls) and blind testing set (including 12 HSCC patients and 16 controls). With WCX2 and IMAC3 protein chips, surface-enhanced laser desorption/ionization (SELDI) was used to analyze the serum protein profiling. 72 samples of the training set were analyzed by a decision tree algorithm to be able to differentiate HSCC patients from controls. Double-blind test was used to determine the sensitivity and specificity of the classification model. Ranging from 2000 - 50000 (M/Z), 11 potential biomarkers on WCX2 and 19 biomarkers on IMAC3 protein chip could differentiate HSCC patients from the control set (P < 10(-5)). Among them 4 candidate protein peaks with the m/z values of 7796, 4216, 5927, and 5361 were selected to be used to establish a predictive model by Biomarker Pattern Software. The model separated effectively the HSCC samples from the control samples, achieving a sensitivity of 94.44%, and a specificity of 88.89%. An accuracy of 85.71% (24/28), sensitivity of 91.67% (11/12), specificity of 81.25% (13/16), positive predictive value of 78.57%% (11/14), and negative predictive value of 92.85% (13/14) were validated in the double-blind testing set. The SELDI-TOF-MS Protein Chip combined with artificial intelligence classification algorithm helps find serum proteome biomarkers and establish predictive model for early diagnosis of HSCC. This technique has potential for the development of a screening test for the detection of HSCC.

Surface‐enhanced laser desorption/ionization time‐of‐flight mass spectrometry serum protein profiling to identify nasopharyngeal carcinoma

Diagnosis of nasopharyngeal carcinoma (NPC) at an early disease stage is important for successful treatment and improving the outcome of patients. The use of serum protein profiles and a classification tree algorithm were explored to distinguish NPC from noncancer. Serum samples were applied to metal affinity protein chips to generate mass spectra by surface-enhanced laser desorption/ionization time-of-flight mass spectrometry (SELDI-TOF-MS). Protein peak identification and clustering were performed using the Biomarker Wizard software. Proteomic spectra of serum samples from 50 NPC patients and 54 noncancer controls were used as a training set and a classification tree with 6 distinct protein masses was generated by using Biomarker Pattern software. The validity of the classification tree was then challenged with a blind test set including another 20 NPC patients and 25 noncancer controls. The software identified an average of 93 mass peaks/spectrum and 6 of the identified peaks were used to construct the classification tree. The classification tree correctly determined 83% (123 of 149) of the test samples with 83% (58 of 70) of the NPC samples and 82% (65 of 79) of the noncancer samples. In a combination of the serum protein profiles with Epstein-Barr (EBV) nuclear antigen 1 (EBNA1 IgA) test, the diagnostic sensitivity and specificity were increased to 99% and 96%, respectively. The results suggest that SELDI-TOF-MS serum protein profiles could discriminate NPC from noncancer. The combination of serum protein profiles with an EBV antibody serology test could further improve the accuracy of NPC screening.

Length-dependent prediction of protein intrinsic disorder.

BackgroundDue to the functional importance of intrinsically disordered proteins or protein regions, prediction of intrinsic protein disorder from amino acid sequence has become an area of active research as witnessed in the 6th experiment on Critical Assessment of Techniques for Protein Structure Prediction (CASP6). Since the initial work by Romero et al. (Identifying disordered regions in proteins from amino acid sequences, IEEE Int. Conf. Neural Netw., 1997), our group has developed several predictors optimized for long disordered regions (>30 residues) with prediction accuracy exceeding 85%. However, these predictors are less successful on short disordered regions (≤30 residues). A probable cause is a length-dependent amino acid compositions and sequence properties of disordered regions.ResultsWe proposed two new predictor models, VSL2-M1 and VSL2-M2, to address this length-dependency problem in prediction of intrinsic protein disorder. These two predictors are similar to the original VSL1 predictor used in the CASP6 experiment. In both models, two specialized predictors were first built and optimized for short (≤30 residues) and long disordered regions (>30 residues), respectively. A meta predictor was then trained to integrate the specialized predictors into the final predictor model. As the 10-fold cross-validation results showed, the VSL2 predictors achieved well-balanced prediction accuracies of 81% on both short and long disordered regions. Comparisons over the VSL2 training dataset via 10-fold cross-validation and a blind-test set of unrelated recent PDB chains indicated that VSL2 predictors were significantly more accurate than several existing predictors of intrinsic protein disorder.ConclusionThe VSL2 predictors are applicable to disordered regions of any length and can accurately identify the short disordered regions that are often misclassified by our previous disorder predictors. The success of the VSL2 predictors further confirmed the previously observed differences in amino acid compositions and sequence properties between short and long disordered regions, and justified our approaches for modelling short and long disordered regions separately. The VSL2 predictors are freely accessible for non-commercial use at

Biomarker discovery for ovarian cancer using SELDI-TOF-MS

Objectives. The purpose of this study is to discover potential biomarkers for the detection and monitoring of adjuvant chemotherapy for ovarian cancer. Methods. Serum samples from ovarian cancers and non-cancer controls were analyzed using surface-enhanced laser desorption/ionization time-of-flight mass spectrometry (SELDI-TOF-MS). To discover the possible diagnostic biomarker for ovarian cancer, a preliminary training set of spectra derived from 31 primary ovarian cancer patients, 16 patients with benign ovarian diseases, and 25 healthy women was used to develop a proteomic model that discriminated cancer from non-cancer effectively. A blind test set, including 43 new cases, was used to validate the sensitivity and specificity of this multivariate model. To explore treatment-induced serum protein change, the protein profiles generated from 16 postoperative patients before chemotherapy are compared with those obtained after chemotherapy. Results. A Four-peak model was established in the training set that discriminated cancer from non-cancer with sensitivity of 90.8% and specificity of 93.5%. A sensitivity of 87.0% and a specificity of 95.0% for the blind test were obtained, compared with 60.7%, 55% for CA125 for the same samples. These 4 markers performed significantly better than the current standard marker, CA125 ( P < 0.05). One protein peak (mass/charge ratio [ m/ z], 4475) was identified in 12 of 16 (75%) postoperative patients after chemotherapy, but was absent before chemotherapy. Conclusion. The proteins represented by these peaks are candidate biomarkers for ovarian cancer diagnosis and/or monitoring treatment response.

Detection of hypopharyngeal squamous cell carcinoma using serum proteomics

Conclusions. The combination of surface-enhanced laser desorption/ionization (SELDI) with bioinformatics tools could help find serum proteome biomarkers and establish a predictive model for early detection of hypopharyngeal squamous cell carcinoma (HSCC). Objectives. Proteomic profiling of serum using surface-enhanced laser desorption/ionization time-of-flight mass spectrometry (SELDI-TOF-MS) is an emerging technique to identify new biomarkers in biological fluids and to establish clinically useful diagnostic computational models. We used it to find new potential biomarkers and to establish a predictive model for early detection of HSCC. Materials and methods. One hundred serum samples including 48 from HSCC patients and 52 from normal controls which were divided into a training set and a blind testing set were treated on WCX2 and IMAC3 protein chip, and serum protein or peptide patterns were detected by SELDI-TOF-MS. The data of spectra were analyzed by Biomarker Wizard software to screen serum proteome biomarkers of HSCC. A decision tree classification algorithm and blind validation were determined by Biomarker Pattern Software (BPS). Results. Ranging from 2 to 30 kDa, 45 potential biomarkers could differentiate HSCC patients from normal controls (p < 0.05). Among them four candidate protein peaks with m/z values of 7796, 4216, 5927, and 5361Da were selected to establish a predictive model by BPS with sensitivity of 94% and specificity of 89%. A sensitivity of 92% and specificity of 82% were validated in the blind testing set.

Serum Protein Profiling Distinguishes BCR-ABL+ CML from Normal and Neutrophilia Cases.

We sought to identify additional biomarkers for chronic myelogenous leukaemia (CML) that could be an aid to early diagnosis and also yield novel antigens for immunotherapy. To this end, we screened patient serum samples at presentation against hematologically normal controls as well as patients with neutrophilia using Surface Enhanced Laser Desorption/Ionization technology (SELDI; Ciphergen ProteinChip series 4000). A total of 84 retrospective and prospective serum samples were analysed: presentation −28, reactive neutrophilia (>15x109 neutrophils/L) − 24 and hematopoietic normal controls − 33. Patients were initially screened by routine cytogenetics and in some cases with qPCR for the BCR-ABL breakpoint. The sera samples were evaluated on 4 different array surfaces and the Immobilised Metal Affinity (IMAC) array was chosen as it bound serum proteins that distinguished CML from normal controls. As little as 1 μl serum was sufficient for each analysis. Biomarker artefacts due to variations in sample collection procedures were ruled out by analysing sera (n=4) from each group at the time of collection and 3 and 6 hours post collection. There were no significant differences in any of the biomarkers at any of the time points. The spectrum of proteins obtained from each of the 84 serum samples was averaged from duplicate runs per experiment. Using the Ciphergen Express program, a panel of 5 proteins were significantly differentially expressed in CML versus the reactive neutrophilia and normal hematopoietic controls (p<0.001). These proteins were identified by a combination of purification techniques using Q HyperD F columns, desalting using reverse phase C-18 beads and isolating the biomarker by 1D-SDS PAGE. The biomarkers were identified by peptide mass fingerprinting and confirmed by Tandem MS sequencing. These wereAlbumin fragment − 2.8Kd (p< 3.5 x 10−5, ROC=0.78),Fibrinogen fragments − 5.3Kd (p< 6.25 x 10−10, ROC=0.07) and 5.9Kd (p< 9.6 x 10−8, ROC=0.14),Complement 3a precursor fragment − 8.9Kd (p< 0.0015, ROC= 0.70),Platelet basic protein precursor − 10.2Kd (p< 1.5 x 10−4, ROC=0.73) andLysozyme − 14.6Kd (p=0, ROC=0.92).Biomarkers 3, 4 and 5 were also verified by antibody capture experiments using NP-20 arrays. In a blinded test set of sera, CML, normal and neutrophilia samples were correctly classified 27/28 (96%), 32/32 (100%), 20/24 (83%) respectively using a combination of the 5.3Kd, 10.2 Kd and the 14.6 Kd markers (Biomaker Pattern software). The algorithm correctly classified 21 new samples as CML (7/8) and control (10/13). The 1/8 CML was misclassified for technical reasons. Therefore, a small number of serum biomarkers in as little as 1 μl serum can be used to distinguish between patients with CML and neutrophilia or hematopoietic normal controls. Similar analyses may be applicable to other more heterogeneous hematological malignancies.

A NIR data set is the object of a chemometric contest at ‘Chimiométrie 2004’

The organisation committee of the symposium 'Chimiométrie 2004' ( http://www.chimiometrie.org/) held in Paris (30th November and 1st December) proposed on their web site a NIR data set. This data set contained 194 spectra of minced meat taken on an Infratec-Tecator instrument between 850 and 1050 nm. The participants were asked to present during the conference their own approaches to calibrate and predict two independent and blind test sets. The committee received nine answers and this paper summarizes the different ways the data were treated by the participants and the proposed approach of the authors. The conference session was quite interesting and considered valuable for a publication by the committee and the participants. The prediction performances expressed by the RMSEP vary within a ratio of 10 between the extremes values. Due to the non-linearity, methods based on classification gave the best prediction models. A simple ANN model gave the best results.

Blood-based biomarkers for detecting mild osteoarthritis in the human knee

This study was designed to test the utility of a blood-based approach to identify mild osteoarthritis (OA) of the knee. Blood samples were drawn from 161 subjects, including 85 subjects with arthroscopically diagnosed mild OA of the knee and 76 controls. Following RNA isolation, an in-house custom cDNA microarray was used to screen for differentially expressed genes. A subset of selected genes was then tested using real-time RT-PCR. Logistic regression analysis was used to evaluate linear combinations of the biomarkers and receiver operating characteristic curve analysis was used to assess the discriminatory power of the combinations. Genes differentially expressed (3543 genes) between mild knee OA and control samples were identified through microarray analysis. Subsequent real-time RT-PCR verification identified six genes significantly down-regulated in mild OA: heat shock 90kDa protein 1, alpha; inhibitor of kappa light polypeptide gene enhancer in B-cells, kinase complex-associated protein; interleukin 13 receptor, alpha 1; laminin, gamma 1; platelet factor 4 (also known as chemokine (C-X-C motif) ligand 4) and tumor necrosis factor, alpha-induced protein 6. Logistic regression analysis identified linear combinations of nine genes--the above six genes, early growth response 1; alpha glucosidase II alpha subunit; and v-maf musculoaponeurotic fibrosarcoma oncogene homolog B (avian)--as discriminatory between subjects with mild OA and controls, with a sensitivity of 86% and specificity of 83% in a training set of 78 samples. The optimal biomarker combinations were then evaluated using a blind test set (67 subjects) which showed 72% sensitivity and 66% specificity. Linear combinations of blood RNA biomarkers offer a substantial improvement over currently available diagnostic tools for mild OA. Blood-derived RNA biomarkers may be of significant clinical value for the diagnosis of early, asymptomatic OA of the knee.

Application of serum SELDI proteomic patterns in diagnosis of lung cancer

BackgroundCurrently, no satisfactory biomarkers are available to screen for lung cancer. Surface-Enhanced Laser Desorption/ionization Time-of- Flight Mass Spectrometry ProteinChip system (SELDI-TOF-MS) is one of the currently used techniques to identify biomarkers for cancers. The aim of this study is to explore the application of serum SELDI proteomic patterns to distinguish lung cancer patients from healthy individuals.MethodsA total of 208 serum samples, including 158 lung cancer patients and 50 healthy individuals, were randomly divided into a training set (including 11 sera from patients with stages I/II lung cancer, 63 from patients with stages III/IV lung cancer and 20 from healthy controls) and a blinded test set (including 43 sera from patients with stages I/II lung cancer, 41 from patients with stages III/IV lung cancer and 30 from healthy controls). All samples were analyzed by SELDI technology. The spectra were generated on weak cation exchange (WCX2) chips, and protein peaks clustering and classification analyses were made using Ciphergen Biomarker Wizard and Biomarker Pattern software, respectively. We additionally determined Cyfra21-1 and NSE in the 208 serum samples included in this study using an electrochemiluminescent immunoassay.ResultsFive protein peaks at 11493, 6429, 8245, 5335 and 2538 Da were automatically chosen as a biomarker pattern in the training set. When the SELDI marker pattern was tested with the blinded test set, it yielded a sensitivity of 86.9%, a specificity of 80.0% and a positive predictive value of 92.4%. The sensitivities provided by Cyfra21-1 and NSE used individually or in combination were significantly lower than that of the SELDI marker pattern (P < 0.005 or 0.05, respectively). Based on the results of the test set, we found that the SELDI marker pattern showed a sensitivity of 91.4% in the detection of non-small cell lung cancers (NSCLC), which was significantly higher than that in the detection of small cell lung cancers (P < 0.05); The pattern also had a sensitivity of 79.1% in the detection of lung cancers in stages I/II.ConclusionThese results suggest that serum SELDI protein profiling can distinguish lung cancer patients, especially NSCLC patients, from normal subjects with relatively high sensitivity and specificity, and the SELDI-TOF-MS is a potential tool for the screening of lung cancer.

Study of serum proteome biomarkers with relation to the formation of portal vein tumor thrombi in hepatocellular carcinoma patients

To screen serum proteome biomarkers and establish predictive model with relation to the formation of portal vein tumor thrombi (PVTT) in hepatocellular carcinoma (HCC) patients. Serum samples were collected from 135 HCC patients, which were divided, into training set (including 33 HCC patients with PVTT and 62 HCC patients without PVTT) and blind testing set (including 18 HCC patients with PVTT and 22 HCC patients without PVTT). Special serum protein or peptide pattern was determined by SELDI-TOF-MS measurement after treating the sample onto WCX2 protein chip for each case. The obtained data were analyzed by BioMarker Wizard software to screen serum proteome biomarkers with relation to the formation of PVTT, while decision tree classification algorithm and blind validation were determined by Biomarker Patterns Software. Ranging from 1100 to 30 000 at the m/z value, 100 protein features were detected in the serum protein pattern stably. Among them, 6 protein peaks with the m/z value of 3478, 1314, 1744, 1725, 2022 and 3380 were upregulated, 10 proteins peaks with the m/z value of 8901, 9353, 9415, 8773, 2766, 2745, 8697, 7773, 8569 and 1373 were downregulated respectively in the group of HCC with PVTT. The 7 candidate protein peaks with the m/z value of 3478, 2022, 8901, 9415, 8773, 2766 and 2745 were selected to establish predictive model by BPS with a sensitivity of 75.8% (25/33) and specificity of 82.3% (51/62). An accuracy of 87.5% (35/40), sensitivity of 100% (18/18), specificity of 77.3% (17/22), positive predictive value of 78.3% (18/23), and negative predictive value of 100% (17/17) were validated in blind testing set. Sixteen candidate proteome biomarkers may be related with the formation of PVTT in HCC patients. Decision tree classification algorithm may have great clinical significance in predicting the formation of PVTT.

Serum protein fingerprinting coupled with artificial neural network distinguishes glioma from healthy population and brain benign tumo

To screen and evaluate protein biomarkers for the detection of gliomas (Astrocytoma grade Ⅰ-Ⅳ) from healthy individuals and gliomas from brain benign tumors by using surface enhanced laser desorption/ionization time of flight mass spectrometry (SELDI-TOF-MS) coupled with an artificial neural network (ANN) algorithm. SELDI-TOF-MS protein fingerprinting of serum from 105 brain tumor patients and healthy individuals, included 28 patients with glioma (Astrocytoma Ⅰ-Ⅳ), 37 patients with brain benign tumor, and 40 age-matched healthy individuals. Two thirds of the total samples of every compared pair as training set were used to set up discriminating patterns, and one third of total samples of every compared pair as test set were used to cross-validate; simultaneously, discriminate-cluster analysis derived SPSS 10.0 software was used to compare Astrocytoma grade Ⅰ-Ⅱ with grade Ⅲ-Ⅳ ones. An accuracy of 95.7%, sensitivity of 88.9%, specificity of 100%, positive predictive value of 90% and negative predictive value of 100% were obtained in a blinded test set comparing gliomas patients with healthy individuals; an accuracy of 86.4%, sensitivity of 88.9%, specificity of 84.6%, positive predictive value of 90% and negative predictive value of 85.7% were obtained when patient's gliomas was compared with benign brain tumor. Total accuracy of 85.7%, accuracy of grade Ⅰ-Ⅱ Astrocytoma was 86.7%, accuracy ofⅢ-Ⅳ Astrocytoma was 84.6% were obtained when grade Ⅰ-Ⅱ Astrocytoma was compared with grade Ⅲ-Ⅳ ones (discriminant analysis). SELDI-TOF-MS combined with bioinformatics tools, could greatly facilitate the discovery of better biomarkers. The high sensitivity and specificity achieved by the use of selected biomarkers showed great potential application for the discrimination of gliomas patients from healthy individuals and glioma from brain benign tumors.

Serum protein fingerprinting coupled with artificial neural network distinguishes glioma from healthy population or brain benign tumor

To screen and evaluate protein biomarkers for the detection of gliomas (Astrocytoma grade I-IV) from healthy individuals and gliomas from brain benign tumors by using surface enhanced laser desorption/ionization time of flight mass spectrometry (SELDI-TOF-MS) coupled with an artificial neural network (ANN) algorithm. SELDI-TOF-MS protein fingerprinting of serum from 105 brain tumor patients and healthy individuals, included 28 patients with glioma (Astrocytoma I-IV), 37 patients with brain benign tumor, and 40 age-matched healthy individuals. Two thirds of the total samples of every compared pair as training set were used to set up discriminating patterns, and one third of total samples of every compared pair as test set were used to cross-validate; simultaneously, discriminate-cluster analysis derived SPSS 10.0 software was used to compare Astrocytoma grade I-II with grade III-IV ones. An accuracy of 95.7%, sensitivity of 88.9%, specificity of 100%, positive predictive value of 90% and negative predictive value of 100% were obtained in a blinded test set comparing gliomas patients with healthy individuals; an accuracy of 86.4%, sensitivity of 88.9%, specificity of 84.6%, positive predictive value of 90% and negative predictive value of 85.7% were obtained when patient's gliomas was compared with benign brain tumor. Total accuracy of 85.7%, accuracy of grade I-II Astrocytoma was 86.7%, accuracy of III-IV Astrocytoma was 84.6% were obtained when grade I-II Astrocytoma was compared with grade III-IV ones (discriminant analysis). SELDI-TOF-MS combined with bioinformatics tools, could greatly facilitate the discovery of better biomarkers. The high sensitivity and specificity achieved by the use of selected biomarkers showed great potential application for the discrimination of gliomas patients from healthy individuals and gliomas from brain benign tumors.

A Pilot Application of SELDI Mass Spectrometry To Study Proteomic Patterns in Plasma from Patients with T-Cell Large Granular Lymphocyte Leukemia.

T-cell large granular lymphocyte (T-LGL) leukemia is a rare chronic lymphoproliferative disorder of cytotoxic T cells of unknown etiology. Chronic antigenic stimulation of T-lymphocytes in autoimmune diseases or viral infection has been postulated to play a role in the pathogenesis of this condition. Surface-enhanced laser desorption/ionization (SELDI) mass spectrometry of plasma and serum specimens was applied as a useful screening method to determine pathophysiologic protein/peptide markers, which may be useful in the diagnosis of occult malignancies. In contrast to traditional diagnostic principles, the SELDI method enables screening of protein patterns without knowing the identity of targets. Patterns of up- and down-modulation of peptides and proteins (resulting from the presence of abnormal products of diseased cells, pathogens or immunogenetic factors) may signal various clinical features and diagnostic clues. We used SELDI mass spectrometry to study proteomic patterns in the plasma of patients with T-LGL leukemia. After initial screening and optimization we selected a hydrophobic H4 ProteinChip surface for further systematic screening. Deviation of mass values and peak intensities were <0.5% and <20%, respectively.Reproducibility was monitored by random duplication of samples with an average of 70 peaks per spectrum (mass/charge ratio from 1500–50,000 Da). Our study included plasma samples from 28 LGL patients, 9 patients with idiopathic neutropenia (without the presence of LGL clone) and 36 healthy blood donors. The diagnosis of T-LGL leukemia included the presence of monoclonal TCR rearrangement and flow cytometric detection of abnormal CD8+CD3+CD57+ cell population. Spectra were analyzed using Biomarker Patterns Software, which allows for internal cross validation for the prediction of the accuracy in a blinded test set. To separate LGL patients from patients with chronic neutropenia, intensity cut-off for 2050 Da peak was found in 22/28 LGL patients, resulting in 100% specificity and 67% sensitivity. Based on the intensities of 3 mass peaks (2274, 5870, 2660 Da), the sensitivity and specificity of LGL distinction from healthy controls was 96% and 61%, respectively. By direct digestion with carboxypeptidase Y on the ProteinChip surface, the C-terminal sequence of the 2050 Da mass peak has been identified. From the sequencing analysis and the search in the ExPASy protein database, we concluded that the 2050 Da mass peak corresponded to the fragment of complement C3 precursor. Additional 2 kDa mass peaks may represent various other C3 fragments. Interestingly, the intensity of the mass peak corresponding to the C3 fragments decreased with the application of immunosuppressive therapy when plasma samples from 3 LGL patients were serially analyzed. The pathophysiologic significance of the identified peak remains unclear, but increased expression of the C3 receptor on LGL cells has been reported. Our data in LGL leukemia indicate that SELDI mass spectrometry can be used to identify immune biomarkers for malignant lymphoproliferations.

Tree analysis of mass spectral urine profiles discriminates transitional cell carcinoma of the bladder from noncancer patient

Background: Recent advances in proteomic profiling technologies, such as surface-enhanced laser desorption/ionization mass spectrometry (SELDI), have allowed preliminary profiling and identification of tumor markers in biological fluids in several cancer types and establishment of clinically useful diagnostic computational models. We developed a bioinformatics tool and used it to identify proteomic patterns in urine that distinguish transitional cell carcinoma (TCC) from noncancer. Methods: Proteomic spectra were generated by mass spectroscopy (surface-enhanced laser desorption and ionization). A preliminary “training” set of spectra derived from analysis of urine from 46 TCC patients, 32 patients with benign urogenital diseases (BUD), and 40 age-matched unaffected healthy men were used to train and develop a decision tree classification algorithm that identified a fine-protein mass pattern that discriminated cancer from noncancer effectively. A blinded test set, including 38 new cases, was used to determine the sensitivity and specificity of the classification system. Results: The algorithm identified a cluster pattern that, in the training set, segregated cancer from noncancer with sensitivity of 84.8% and specificity of 91.7%. The discriminatory pattern correctly identified. A sensitivity of 93.3% and a specificity of 87.0% for the blinded test were obtained when comparing the TCC vs. noncancer. Conclusions: These findings justify a prospective population-based assessment of proteomic pattern technology as a screening tool for bladder cancer in high-risk and general populations.

Determination of geographical origin of olive oils using 13C nuclear magnetic resonance spectroscopy. I - Classification of olive oils of the Puglia region with denomination of protected origin.

13C nuclear magnetic resonance spectroscopy was used to classify olive oils from the three production areas of the Puglia region labeled with the "denomination of protected origin" (DPO) Terra di Bari, Colline di Brindisi, and Dauno. High resolution (13)C spectra of 173 olive oil samples were measured, and the intensity data of triacylglycerol resonances were processed by using linear discriminant analysis, which was carried out stepwise for variable selection. The olive oil samples from the DPOs Colline di Brindisi and Terra di Bari were 90% correctly classified, whereas only 74% of "Dauno" DPO oils were classified in the true group. The performance of the discriminant model was verified by applying the cross-validation procedure based on the "leave one out" formalism. The discriminant model was evaluated against a blind test set of olive oils from the three DPO areas. All the oils used for the purpose were correctly assigned to their respective groups, with the exception of the Dauno oil samples based on the Coratina cv. They were misclassified as Terra di Bari oils because of a common monovarietal composition.

Neocognitron for handwritten digit recognition

The author previously proposed a neural network model neocognitron for robust visual pattern recognition. This paper proposes an improved version of the neocognitron and demonstrates its ability using a large database of handwritten digits (ETL1). To improve the recognition rate of the neocognitron, several modifications have been applied: such as, the inhibitory surround in the connections from S-cells to C-cells, contrast-extracting layer between input and edge-extracting layers, self-organization of line-extracting cells, supervised competitive learning at the highest stage, staggered arrangement of S- and C-cells, and so on. These modifications allowed the removal of accessory circuits that were appended to the previous versions, resulting in an improvement of recognition rate as well as simplification of the network architecture. The recognition rate varies depending on the number of training patterns. When we used 3000 digits (300 patterns for each digit) for the learning, for example, the recognition rate was 98.6% for a blind test set (3000 digits), and 100% for the training set.

Wavelet change-point prediction of transmembrane proteins

A non-parametric method, based on a wavelet data-dependent threshold technique for change-point analysis, is applied to predict location and topology of helices in transmembrane proteins. A new propensity scale generated from a transmembrane helix database is proposed. We show that wavelet change-point performs well for smoothing hydropathy and transmembrane profiles generated using different scales. We investigate which wavelet bases and threshold functions are overall most appropriate to detect transmembrane segments. Prediction accuracy is based on the analysis of two data sets used as standard benchmarks for transmembrane prediction algorithms. The analysis of a test set of 83 proteins results in accuracy per segment equal to 98.2%; the analysis of a 48 proteins blind-test set, i.e. containing proteins not used to generate the propensity scales, results in accuracy per segment equal to 97.4%. We believe that this method can also be applied to the detection of boundaries of other patterns such as G + Cisochores and dot-plots. The transmembrane database, TMALN and source code are available upon request from the authors.

The Evolution of evaluation: Lessons from the Message Understanding Conferences

The Message Understanding Conferences (MUCs) represent one of the earliest and longest running efforts to evaluate language understanding technology. This article reviews the history of the MUCs and their evolution towards the use of common training and blind test sets, automated scoring, task decomposition into modular building blocks and tools for portability across languages and applications. Now that evaluation has become an accepted part of the developer's toolkit, it is important to understand the interplay between evaluation methods and the state of research. MUC was successful in generating excitement about text processing problems and in attracting talented researchers to the area. It also provided a functional decomposition of the information extraction problem into a series of simpler problems, thus allowing researchers to demonstrate successful systems and to spin off commercial products. However, the ultimate goal of accurate information extraction has been elusive; systems have become faster and cheaper to build, the evaluations have become harder, but overall accuracy in information extraction has improved only modestly. The MUC experience contrasts with experiences in other evaluations. For example, the spoken evaluation in the Air Travel Information System (ATIS) has shown dramatic improvement in error rate over time, but those evaluations were limited to a single domain and the metrics did not address interaction, even though real-time interactive systems were available. Looking across the history of MUC in the context of related evaluations, we can draw important lessons about the need for evaluation to evolve with the technology it evaluates, to balance costs against benefits and to weigh the divergent needs of the multiple stake-holders— developers, funders and users—in order to provide continuity while also providing the next set of challenges to the research community.

An algorithm for pronominal anaphora resolution

This paper presents an algorithm for identifying the noun phrase antecedents of third person pronouns and lexical anaphors (reflexives and reciprocals). The algorithm applies to the syntactic representations generated by McCord's Slot Grammar parser and relies on salience measures derived from syntactic structure and a simple dynamic model of attentional state. Like the parser, the algorithm is implemented in Prolog. The authors have tested it extensively on computer manual texts and conducted a blind test on manual text containing 360 pronoun occurrences. The algorithm successfully identifies the antecedent of the pronoun for 86% of these pronoun occurrences. The relative contributions of the algorithm's components to its overall success rate in this blind test are examined. Experiments were conducted with an enhancement of the algorithm that contributes statistically modelled information concerning semantic and real-world relations to the algorithm's decision procedure. Interestingly, this enhancement only marginally improves the algorithm's performance (by 2%). The algorithm is compared with other approaches to anaphora resolution that have been proposed in the literature. In particular, the search procedure of Hobbs' algorithm was implemented in the Slot Grammar framework and applied to the sentences in teh blind test set. The authors' algorithm achieves a higher rate of success (4%) than Hobbs' algorithm. The relation of the algorithm to the centering approach is discussed, as well as to models of anaphora resolution that invoke a variety of informational factors in ranking antecedent candidates.