Statistical Analysis and Multimodal Classification on Noisy Eye Tracker and Application Log Data of Children with Autism and ADHD

Emotion recognition behavior and performance may vary between people with major neurodevelopmental disorders such as Autism Spectrum Disorder (ASD), Attention Deficit Hyperactivity Disorder (ADHD) and control groups. It is crucial to identify these differences for early diagnosis and individual treatment purposes. This study represents a methodology by using statistical data analysis and machine learning to provide help to psychiatrists and therapists on the diagnosis and individualized treatment of participants with ASD and ADHD. In this paper we propose an emotion recognition experiment environment and collect eye tracker fixation data together with the application log data (APL). In order to detect the diagnosis of the participant we used classification algorithms with the Tomek links noise removing method. The highest classification accuracy results were reported as 86.36% for ASD vs. Control, 81.82% for ADHD vs. Control and 70.83% for ASD vs. ADHD. This study provides evidence that fixation and APL data have distinguishing features for the diagnosis of ASD and ADHD.     

No free lunch for early stopping

We show that with a uniform prior on models having the same training error, early stopping at some fixed training error above the training error minimum results in an increase in the expected generalization error.     

Identification of insect damaged wheat kernels using transmittance images

Transmittance images of wheat kernels are used to classify insect damaged and undamaged wheat kernels. The colour histograms of pixel intensities of the wheat images were used as the feature vectors. Combination of the linear model and a radial basis function network, in a committee, resulted in a false positive rate of 0.1 at the true positive rate of 0.8 and an area under the receiver operating characteristics curve of 0.92.     

Cross-class inhibition of the cysteine proteinases cathepsins K, L, and S by the serpin squamous cell carcinoma antigen 1: a kinetic analysis

The human squamous cell carcinoma antigens (SCCA) 1 and 2 are tandemly arrayed genes that encode two high-molecular-weight serine proteinase inhibitors (serpins). Although these proteins are 92% identical, differences in their reactive site loops suggest that they inhibit different types of proteinases. Our previous studies show that SCCA2 inhibits chymotrypsin-like serine proteinases [Schick et al. (1997) J. Biol. Chem. 272, 1849-1855]. We now show that, unlike SCCA2, SCCA1 lacks inhibitory activity against any of the more common types of serine proteinases but is a potent cross-class inhibitor of the archetypal lysosomal cysteine proteinases cathepsins K, L, and S. Kinetic analysis revealed that SCCA1 interacted with cathepsins K, L, and S at 1:1 stoichiometry and with second-order rate constants >/= 1 x 10(5) M-1 s-1. These rate constants were comparable to those obtained with the prototypical physiological cysteine proteinase inhibitor, cystatin C. Also relative to cystatin C, SCCA1 was a more potent inhibitor of cathepsin K-mediated elastolytic activity by forming longer lived inhibitor-proteinase complexes. The t1/2 of SCCA1-cathepsin S complexes was >1155 min, whereas that of cystatin C-cathepsin complexes was 55 min. Cleavage between the Gly and Ser residues of the reactive site loop and detection of a stable SCCA1-cathepsin S complex by sodium dodecyl sulfate-polyacrylamide gel electrophoresis suggested that the serpin interacted with the cysteine proteinase in a manner similar to that observed for typical serpin-serine proteinase interactions. These data suggest that, contingent upon their reactive site loop sequences, mammalian serpins, in general, utilize their dynamic tertiary structure to trap proteinases from more than one mechanistic class and that SCCA1, in particular, may be involved in a novel inhibitory pathway aimed at regulating a powerful array of lysosomal cysteine proteinases.     

SCCA1 and SCCA2 are proteinase inhibitors that map to the serpin cluster at 18q21.3

Vascular endothelial growth factor (VEGF) is an essential molecule in the development and formation of mammalian blood vessels in health and disease. VEGF is also increasingly implicated in other biological processes including renal development and pathophysiology. The biological activities of VEGF in vivo and in its target cells in culture are mediated through two receptor protein tyrosine kinases, KDR/Flk-1 and Flt-1. KDR/Flk-1 is able to mediate the tyrosine phosphorylation of several cellular components as well as the generation of second messengers. Recent findings have revealed novel signaling mechanisms which may mediate the biological actions of VEGF. In contrast, the signal transduction mechanisms triggered by Flt-1 remain largely unknown.     

Improving Performance Prediction on Education Data with Noise and Class Imbalance

This paper proposes to apply machine learning techniques to predict students’ performance on two real-world educational data-sets. The first data-set is used to predict the response of students with autism while they learn a specific task, whereas the second one is used to predict students’ failure at a secondary school. The two data-sets suffer from two major problems that can negatively impact the ability of classification models to predict the correct label; class imbalance and class noise. A series of experiments have been carried out to improve the quality of training data, and hence improve prediction results. In this paper, we propose two noise filter methods to eliminate the noisy instances from the majority class located inside the borderline area. Our methods combine the over-sampling SMOTE technique with the thresholding technique to balance the training data and choose the best boundary between classes. Then we apply a noise detection approach to identify the noisy instances. We have used the two data-sets to assess the efficacy of class-imbalance approaches as well as both proposed methods. Results for different classifiers show that, the AUC scores significantly improved when the two proposed methods combined with existing class-imbalance techniques.     

Anatomical studies of DNA fragmentation in rat brain after systemic kainate administration

Rats treated systemically with kainate develop stereotyped epileptic seizures involving mainly limbic structures that may last for hours. This model of limbic status epilepticus has been widely studied using classical neuropathological techniques. We used in situ nick translation histochemistry to examine patterns of DNA fragmentation in this model. We found a stereotyped and reproducible pattern of neuronal populations that demonstrate evidence of DNA fragmentation from 24 h to one week after kainate treatment. Neither blockade of new protein synthesis nor blockade of the N-methyl-D-aspartate-type glutamate receptors significantly altered this response. Moreover, we saw no evidence of the regular internucleosomal cleavage of DNA that produces a characteristic laddered appearance of 180-200 bp DNA fragments after gel electrophoresis in samples obtained from microdissected affected regions. These studies suggest that DNA fragmentation after systemic kainate-induced seizures is not the result of programmed cell death. This assay may be useful for quantitative testing of both neuroprotective agents and mechanistic hypotheses.     

General sensitivity formulas for lossless two-ports

Simple first-order sensitivity relations, which give the effects of component-value variations, losses and stray elements in both singly and doubly terminated lossless two-ports, are introduced. They provide a generalisation of some earlier results, and also lead to some new formulas. They are also applicable to active-RC and switched-capacitor filters modelled on doubly terminated lossless ladder two-ports.