Page 280 - Artificial Intelligence in the Age of Neural Networks and Brain Computing

4. Multiview Learning in Neuroinformatics 273

(A) (B) (C)

(D)

(E) (F) (G)

FIGURE 13.7
Drug repositioning algorithms can integrate different kinds of information, such as text (A)
genome-wide studies (B) gene expression (C) structural properties (D) adverse events
such as drugs side effects (E) interactiome (F) pathways (G).

structure, and drugs targets. They applied a kernel-based late integration approach
where for each view they constructed a distance matrix and then they combined
these matrices by creating a mean kernel. The first similarity matrix is the correla-
tion between the gene expression patterns; the second depends on how similar the
drugs are with respect to their chemical structure; and the last one is the distance
matrix between drug targets in their proteineprotein interaction network. The
combined matrix was used to train the multiclass SVM classifier in order to
predict therapeutic classes. Their results show a high accuracy in the classification
task that allows for the repositioning of systematically misclassified drugs.

4. MULTIVIEW LEARNING IN NEUROINFORMATICS

4.1 AUTOMATED DIAGNOSIS SUPPORT TOOLS FOR
NEURODEGENERATIVE DISORDERS
Magnetic resonance imaging (MRI) is a noninvasive imaging technique that
exploits the magnetic properties of tissues to describe brain function at high
temporal and spatial resolutions. The application of this method has brought great
advances in neuroscience research, allowing for the investigation of the physio-
logical and pathological mechanisms governing the human brain. Nevertheless,
the knowledge transfer from research studies to clinical practice has been
hindered by the limitations of standard analysis methods. For instance, commonly