Mining Biomedical Literature for Neuroanatomy

Leon FrenchBioinformatics Training Program

Rotation Supervisor: Dr. Paul Pavlidis

Neuroinformatics

Application of informatics technologies to neuroscience study of the nervous system modeling

The Human Connectome: A Structural Description of the Human BrainOlaf Sporns, Guilio Tononi, Rolf Kötter (2005)

“Connection matrix of the human brain” Similar to the “Interactome” Methods:

Tracing Stimulation DT-MRI

Why?

Increased understanding of the structure and function of the brain, connectivity -> function

Once digitized it can be integrated with other databases

Need for accessible connection data: standardized data formats few public databases (CoCoMac, BAMS)

Currently: Expert knowledge Literature searches Atlases or maps – little connectivity data

NeuroNames Brain Hierarchy Nomenclature, hierarchy of terms Contains over 6,500 terms and 550 primary

structures Example: Basal Nucleus

Brain > Forebrain > Telencephalon > Basal Nucleus Synonym: nucleus basalis

Part of UMLS

The Idea – Text Mining

Search all Medline abstracts for all Neuronames, and look for interesting co-occurrences

Text mine abstracts using natural language processing

Infer connections between brain regions Similar past work:

Automatic extraction of biological information from scientific text: protein-protein interactions (Blaschke, et al. 1999)

Abstract Example - RegionsRole of the central nucleus of the amygdala in the control of blood pressure: descending pathways to medullary cardiovascular nuclei., Saha S., 2005

One of the key areas that links psychologically induced stress with the blood pressure-regulatory system is the central nucleus of the amygdala (CeA). This is an integratory forebrain nucleus that receives input from higher centres in the forebrain and has extensive connections with the hypothalamus and the medulla oblongata, areas involved in the regulation of the cardiovascular reflexes.

Abstract Example -VerbsBrain Res Bull. 1985 Mar;14(3):261-72. Afferent and efferent connections of the medial preoptic

area in the rat: a WGA-HRP study.Chiba T, Murata Y.

Afferent and efferent connections of the medial preoptic area including medial preoptic nucleus (MP) and periventricular area at the MP level were examined using WGA-HRP as a marker. Injections were performed by insertion of micropipette containing (1) small amount of HRP powder or (2) dryed HRP solution for 24 to 48 hr until the fixation or for 5 min respectively. Dorsal and ventral approaches of injection micropipettes were performed and the results were compared. Previously reported reciprocal connections with lateral septum, bed nucleus of the stria terminalis, medial amygdaloid nucleus, lateral hypothalamic nucleus, paraventricular hypothalamic nucleus, ventromedial hypothalamic nucleus, arcuate nucleus, supramammillary nucleus, central gray at the mesencephalon, raphe dorsalis, raphe medianus, and lateral parabrachial nucleus have been confirmed. In addition, we found reciprocal connections with septo-hypothalamic nucleus, amygdalo-hipocampal nucleus, subiculum, parafascicular thalamic nucleus, posterior thalamic nucleus at the caudo-ventral subdivision, median preoptic nucleus, lateral preoptic nucleus, anterior hypothalamic nucleus, periventricular area at the caudal hypothalamic level, dorsomedial hypothalamic nucleus, posterior hypothalamic nucleus, dorsal and ventral premammillary nucleus, lateral mammillary nucleus, peripeduncular nucleus, periventricular gray, ventral tegmental area, interpeduncular nucleus, nucleus raphe pontis, nucleus raphe magnus, pedunculo-pontine tegmental nucleus, gigantocellular reticular nucleus and solitary tract nucleus. The areas which had only efferent connections from MP were accumbens, caudate putamen, ventral pallidum, substantia innominata, lateral habenular nucleus, paratenial thalamic nucleus, paraventricular thalamic nucleus, mediodorsal thalamic nucleus, reuniens thalamic nucleus, median eminence, medial mammillary nucleus, subthalamic nucleus, pars compacta of substantia nigra, oculomotor nucleus, red nucleus, laterodorsal tegmental nucleus, reticular tegmental nucleus, cuneiform nucleus, nucleus locus coeruleus, and dorsal motor nucleus of vagus among which substantia innominata and median eminence were previously reported. Efferent connections to the nucleus of Darkschewitsch, interstitial nucleus of Cajal, dorsal tegmental nucleus, ventral tegmental nucleus, vestibular nuclei, nucleus raphe obsculus were very weak or abscent in the ventral approach while they were observed in dorsal approach. Previously reported afferent connections from dorsal tegmental nucleus, cuneiform nucleus, and nucleus locus ceruleus were not detected in this study.(ABSTRACT TRUNCATED AT 400 WORDS)

Early Results

Reduced the 9 million abstracts to 1.25 million that contain at least one Neuroname

Reduced 6000 Neuronames to 2626 that appear in at least one abstract.

Computed covariance matrix of Neuroname counts (co-occurrence) two brain regions vary together in literature hits

~> they are connected

Sub-graph from Covariance Matrix

Future work

Background reading Neuroanatomy Protein-Protein interaction

Create gold standard set of abstracts Natural language processing:

“region a projects to region b” Named entity recognition

Incorporate other concepts – mental processes, symptoms, or genes

Filtering and Visualization of Large Datasets through a Semantic Lens Supervised by Dr. Mark Wilkinson at iCapture

centre Mining biomedical literature for UMLS

concepts Based on work by Benjamin Good, Byron Ku

and Edward Kawas

Current Rotation – Semantic Lensing

Meaningfulperspective

SemanticLens

(ontology)Data

NeuroNames

200 Alzheimerabstracts

Mental Processes

200 Alzheimerabstracts

That’s it so far..

Thanks for your attention

Acknowledgements Pavlidis Lab UBC Bioinformatics Centre (UBiC) Wilkinson Lab at the iCapture Centre CIHR Bioinformatics Training Program Natural Sciences and Engineering Research

Council of Canada

Questions?