Van T. T. Nguyen

Potent and long-acting dimeric inhibitors of influenza virus neuraminidase are effective at a once-weekly dosing regimen.

ABSTRACT Dimeric derivatives (compounds 7 to 9) of the influenza virus neuraminidase inhibitor zanamivir (compound 2), which have linking groups of 14 to 18 atoms in length, are approximately 100-fold more potent inhibitors of influenza virus replication in vitro and in vivo than zanamivir. The observed optimum linker length of 18 to 22 Å, together with observations that the dimers cause aggregation of isolated neuraminidase tetramers and whole virus, indicate that the dimers benefit from multivalent binding via intertetramer and intervirion linkages. The outstanding long-lasting protective activities shown by compounds 8 and 9 in mouse influenza infectivity experiments and the extremely long residence times observed in the lungs of rats suggest that a single low dose of a dimer would provide effective treatment and prophylaxis for influenza virus infections.

Graph-based semi-supervised Support Vector Data Description for novelty detection

Support Vector Data Description (SVDD) is a well-known supervised learning method for novelty detection purpose. For its classification task, SVDD requires a fully-labeled dataset. Nonetheless, contemporary datasets always consist of a collection of labeled data samples jointly a much larger collection of unlabeled ones. This fact impedes the usage of SVDD in the real-world problems. In this paper, we propose to utilize the information implicated in a spectral graph to leverage SVDD in the context of semi-supervised learning. The theory and experiment evidence that the proposed method is able to efficiently employ the information carried in the spectral graph to not only enhance the generalization ability of SVDD but also enforce the cluster assumption which is crucial for a semi-supervised learning method.

Jointly Predicting Affective and Mental Health Scores Using Deep Neural Networks of Visual Cues on the Web

Despite the range of studies examining the relationship between mental health and social media data, not all prior studies have validated the social media markers against “ground truth”, or validated psychiatric information, in general community samples. Instead, researchers have approximated psychiatric diagnosis using user statements such as “I have been diagnosed as X”. Without “ground truth”, the value of predictive algorithms is highly questionable and potentially harmful. In addition, for social media data, whilst linguistic features have been widely identified as strong markers of mental health disorders, little is known about non-textual features on their links with the disorders. The current work is a longitudinal study during which participants’ mental health data, consisting of depression and anxiety scores, were collected fortnightly with a validated, diagnostic, clinical measure. Also, datasets with labels relevant to mental health scores, such as emotional scores, are also employed to improve the performance in prediction of mental health scores. This work introduces a deep neural network-based method integrating sub-networks on predicting affective scores and mental health outcomes from images. Experimental results have shown that in the both predictions of emotion and mental health scores, (1) deep features majorly outperform handcrafted ones and (2) the proposed network achieves better performance compared with separate networks.

Maximal margin learning vector quantisation

Kernel Generalised Learning Vector Quantisation (KGLVQ) was proposed to extend Generalised Learning Vector Quantisation into the kernel feature space to deal with complex class boundaries and thus yielded promising performance for complex classification tasks in pattern recognition. However KGLVQ does not follow the maximal margin principle, which is crucial for kernel-based learning methods. In this paper we propose a maximal margin approach (MLVQ) to the KGLVQ algorithm. MLVQ inherits the merits of KGLVQ and also follows the maximal margin principle to improve the generalisation capability. Experiments performed on the well-known data sets available in UCI repository show promising classification results for the proposed method.