From cortisone to graphene: 60 years of breakthroughs in PubMed publications
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
From cortisone to graphene:
60 years of breakthroughs in PubMed
publications
André Panisson, Marco Quaggiotto
Data Science Laboratory, ISI Foundation, Torino, Italy
This report refers to the methodology used to create the visualization entitled
“From cortisone to graphene: 60 years of breakthroughs in PubMed publica-
tions”, submitted to the Data Visualization Challenge of the 2014 ACM Web
Science Conference.
Biomedical science advances can often be described by brief periods of dis-
covery followed by decades of study and research. This analysis aims to show
an overview of the breakthroughs inferred from the titles of 21.5 millions of
PubMed publications. For this visualization, we built a timeline of terms that
have experienced a sudden increase in popularity in PubMed publication titles.
The position of the terms is related to the years in witch the breakthrough has
occurred, the size and color of the word are indicative of the popularity of the
term. Authors that contributed significantly to the breakthroughs have been
linked to their areas of research.
In order to create the visualization we:
1. Extracted words and bigrams from publication titles
2. Analyzed the occurrence of such terms in the years between 1950 and 2010
3. Detected peaks of activity for each term, identifying the years with high
acceleration in the term usage
4. Filtered out terms with low entropy as they are too generic
5. Identified related authors (with at least 100 publications, of which at least
50 contain the given term in the title)
6. Created a weighted graph with years, terms and authors
7. Used a force-directed layout to position the terms next to their years and
authors next to their related terms
In the next sections, we discuss this methodology in detail.
11 Dataset
PubMed comprises more than 23 million citations for biomedical literature from
MEDLINE, life science journals, and online books. In this work, we used meta-
data for the complete set of all PubMed records, including title, authors, and
year of publication. All data provided originates from NLMs PubMed database
(as downloaded April 24, 2013 from the NLM FTP site) and was retrieved via
the Scholarly Database 1;2 .
The MEDLINE R /PubMed R dataset contains articles starting before 1900,
however the number of publications is very small for each year in the 19th
century (less than 1000 per year). After 1900, the number of publications per
year grows above 1000, but a big increase starts in the 1940s, when the number
of publications jumps from 3357 in 1944 to 17148 in 1945. Figure 1 shows a
timeline with the number of publications in the dataset for each year, starting
from 1900.
0.9M
0.8M
0.7M
Number of publications
0.6M
0.5M
0.4M
0.3M
0.2M
0.1M
0.0M
1900 1920 1940 1960 1980 2000 2020
Year of publication
Figure 1: Number of publications in the MEDLINE R /PubMed R dataset start-
ing from 1900
In order to limit the analysis to years with high number of publications, we
decided to show in the visualization only data from 1950 to 2010.
2 Text analysis
Our method starts with text analysis of the article titles. In order to analyze
the article titles, we use the “bag of words” approach, and we create a matrix
A ∈ RN ×M , where N is the number of articles and M is the number of terms
extracted from the dataset. Each position ai,j represents the number of times
the term j appears in the title of article i.
In order to extract the terms, we use the steps that we show in the following.
We illustrate each step with the example of the title ‘Drugs and the high
cost of health care.’:
(1) choose a list of stop words (common words in English and in article titles):
[is, and, the, of, ..., letter, article, conference, ...]
(2) extract a list of words from the article title: [drugs, and, the, high,
cost, of, health, care]
2(3) lemmatize the words of the list (e.g., drugs becomes drug): [drug, and,
the, high, cost, of, health, care]
(4) remove the stop words and create n-grams with maximum size of 2 by
concatenating words only if they are not separated by a stop word: [drug,
high, cost, health, care, high cost, health care]
For each article title, we apply the above process and add the resulting
terms to a counter. The result is a dictionary that associates each term with
the number of times the term appears in the dataset. We select only the 100
thousand most common terms and proceed to the construction of matrix A,
where the number of terms M is 100 thousand.
3 Time series analysis
Since we want to select terms that could represent discoveries and breakthroughs,
we analyze the time-series of the occurrences of each term over the years.
For each term, we start by creating a time series with the number of term
occurrences in each year. For example, the term hiv exhibits a time series of
occurrences as shown in Figure 2.
8000
7000
6000
nr of ocurrences of hiv
5000
4000
3000
2000
1000
0
1950 1960 1970 1980 1990 2000 2010
year
Figure 2: Time series with number of occurrences of the term hiv from 1950 to
2010.
In this figure, we see the increasing occurrence of the term hiv, but this
time series is affected by the increasing number of articles in the dataset. In
order to remove the effect of increasing number of articles, we normalize the
frequencies by dividing the term occurrence by the sum of all term occurrences
in each year. This gives us the time series with the relative frequency of the
terms, in the following referred as f (t). For the term hiv, we show its f (t) in
Figure 3.
We derive the following time series from f (t):
δ(f (t)) = f (t) − f (t − 1)
2
δ (f (t)) = δ(f (t)) − δ(f (t − 1))
In this example, δ 2 (f (t)) corresponds to the second discrete derivative of
f (t), i.e., the years where the frequency acceleration is at its maximum.
30.012
0.010
relative frequency of hiv
0.008
0.006
0.004
0.002
0.000
1950 1960 1970 1980 1990 2000 2010
year
Figure 3: Time series with relative frequency (f (t)) of the term hiv from 1950
to 2010.
f (t)
ϕ(f (t)) =
f (t − 1) + ε
1−ϕ(f (t))
ω(f (t)) = 1 − e C
g(t) = δ 2 (f (t))δ(f (t))ω(f (t))
ε is a small number used to avoid zero division. The parameter C is used to
give less weight to frequencies that jump from high values to higher values, and
more weight to frequencies that jump from very low values to high values. In
our experiments, we used C = 10. The resulting time series g(t) is given by the
product of three time series: δ 2 (f (t))δ(f (t))ω(f (t)). We multiply δ 2 (f (t)) - the
second discrete derivative of f (t) - by δ(f (t)) - the first discrete derivative of
f (t), since we want the years where there is a high acceleration in the term usage
and also the years where the increase in term usage is positive. We multiply by
ω(f (t)) in order to give less weight to the increases that have a relative small
change from t − 1 to t. Finally, we set all negative values to 0 in order to keep
only positive peaks.
g(t) has peaks in the years where the frequency goes from near zero to a high
value. In order to classify the terms by their frequency jumps, we use a measure
of entropy. In Figure 4, for some terms, we show in blue the initial time series
with the relative frequency -f (t), in red the resulting g(t) for the term, and we
show next to the term name the resulting entropy calculated over g(t).
Using this methodology to analyze the time series of each term, we are able
to extract two types of information:
1. Most of the terms with high entropy (next to 1) are terms that are used in
all years, or that have no visible jump in the frequency. By selecting only
terms with low entropy and high accumulated frequency, we have a good
chance to extract terms that represent discoveries and breakthroughs from
1950 to 2010.
40.09 hiv (0.820) 0.5 graphene (0.251) 0.25 cortisone (0.193)
0.08
0.07 0.4 0.20
0.06
0.05 0.3 0.15
0.04 0.2 0.10
0.03
0.02 0.1 0.05
0.01
0.00 0.0 0.00
1950 1960 1970 1980 1990 2000 2010 1950 1960 1970 1980 1990 2000 2010 1950 1960 1970 1980 1990 2000 2010
0.030 cell (0.999) 0.08 global (0.999) 0.030 variable (0.999)
0.025 0.07 0.025
0.06
0.020 0.05 0.020
0.015 0.04 0.015
0.010 0.03 0.010
0.02
0.005 0.01 0.005
0.000 0.00 0.000
1950 1960 1970 1980 1990 2000 2010 1950 1960 1970 1980 1990 2000 2010 1950 1960 1970 1980 1990 2000 2010
Figure 4: For some terms, we show (i) in blue the time series with the relative
frequency (f (t)), (ii) in red the processed g(t) for the term, and (iii) next to the
term name, the resulting entropy calculated over g(t).
2. Terms with low entropy have high peaks that can be associated to the year
of their frequency increase. The values in g(t) associated to each year can
be used as a weight to associate a given term to the year of its discovery.
3.1 A small validation
One simple way to validate our method for finding the discoveries associated
to each year is to check if terms associated to Nobel Prizes are present in the
extracted terms. The 1950 Nobel Prize in Medicine was given to Edward Calvin
Kendall, Tadeus Reichstein and Philip Showalter Hench “for their discoveries
relating to the hormones of the adrenal cortex, their structure and biological
effects”. The term associated to this discovery is “cortisone”. The 2010 No-
bel Prize in Physics was given to Andre Geim and Konstantin Novoselov “for
groundbreaking experiments regarding the two-dimensional material graphene”,
and the associated term is “graphene”. These two terms motivated the title of
our work: “From cortisone to graphene”. We were able to verify that both terms
are present in the visualization. Other terms like “cell”, “global” and “variable”
have high frequency in the dataset, but since they present low entropy for their
corresponding g(t), they were excluded from the visualization.
4 Author selection
The MEDLINE R /PubMed R dataset contains around 10.7 million author names
associated to the articles comprised between 1950 and 2010. Each author name
is expressed as ‘Surname, Name’ (e.g., ‘Smith, John’) or as ‘Surname, intitals’
(e.g., ‘Smith, J’). From the articles and author names, we can build a matrix
B ∈ RN ×O , where N is the number of articles and O is the number of author
names. Each position bi,j is set to 1 if the author name j is associated to article
i, and is set to 0 otherwise.
5The author name associated to the highest number of articles is ‘Suzuki, T’.
Clearly, this author name represents more than one real author. In fact, the
page of PubMed Advanced Search shows that there are tens of authors with
surname ‘Suzuki’ and initial ‘T’. In order to create a list of authors associated
to each term, we wanted to avoid using author names that do not refer to a
single author. For this, we also used a measure of entropy for each author. The
idea behind it is that author names that are associated to a high number of
terms are most probably different authors, since authors are committed to few
topics. The entropy of each author is calculated using the following equation:
M
X
Hj = − p(i|j) log(p(i|j))
i=1
where p(i|j) is the probability of having a term i given an author j. From
this, we calculate a scaled entropy:
Hj
Ej = 1 −
Hmax
Figure 5 shows an histogram with the distribution of number of authors for
each bin of Ej values and the number of removed author names when selecting
only author names with Ej above a threshold t (Ej > t). We verified that
setting Ej > 0.25, we still keep more than 99.7% of the author names and, at
the same time, we remove all top author names that refer to more than one real
author.
600000 1.0
500000 0.8
percent of authors removed
400000
number of authors
0.6
300000
0.4
200000
100000 0.2
0 0.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
enthropy enthropy threshold
Figure 5: In the left, an histogram showing the distribution of number of author
names for each bin of Ej values (we used 100 bins). In the right, the number
of removed author names when selecting only author names with Ej above an
entropy threshold t (Ej > t).
From the list of author names with Ej > 0.25, we identified only the authors
with least 100 publications. Next, use the document-term matrix A to calculate
the co-occurrence of authors and terms: C = AT B, with C ∈ RM ×O . From
the matrix C we select only the rows that refer to the terms chosen for the
visualization. Finally, for each row, we select only the top 2 authors which used
at least 50 times the given term in their articles.
65 Graph creation and layout
We used the years, terms and authors to create a weighted graph as follows:
• V1 is the set of nodes representing the years, i.e., from 1950 to 2010 (61
nodes).
• V2 is the set of nodes representing the selected terms.
• V3 is the set of nodes representing the selected authors.
• E12 is a set of weighted edges connecting years with terms. The weights
are proportional to the values of the g(t) corresponding to each term.
• E22 is a set of weighted edges connecting terms with terms. The weights
are proportional to the co-occurrence of the terms in article titles.
• E23 is a set of weighted edges connecting terms with authors. The weights
are proportional to the number of times an author used the term in article
titles.
• V is the union of V1 , V2 and V3 .
• E is the union of E12 , E22 and E23 .
G is a graph with nodes V and edges E. This graph was exported to a file
and imported to a graph exploration tool (in this case, Gephi 3 ). Finally, we set
the year nodes to fixed positions, and used two different layout algorithms to
distribute spatially the term and author nodes: Force Atlas and Label Adjust.
The size and color of term nodes were set proportionally to the term frequency,
as shown in Figure 6.
Hardie, D Grahame Viollet, Benoit
mitogen activate
cyclin
fluorescent protein factor kappab
activate protein
overexpression knockout matrix metalloproteinase Li, Wen-Jun O'Hair, Richard A J
Minamino, N Davis, M M
tumor suppressor
Devroey, P
alpha1
Van Steirteghem, A
intellectual disability lamivudine soil gas phase
vibrational
natriuretic peptide bariatric
Ginsberg, M H
cell receptor
factor alpha integrin
Baird, A
asymmetric Shibasaki, Masakatsu
dendritic
Zhou, Qixing
intracytoplasmic rna interference nanocomposite
recombinant human exon cyclin dependent endovascular treatment catenin sediment
von Boehmer, H
continuous ambulatory Emmrich, F Matsumoto, Kunio Presta, M
Humphries, M J Umezawa, Kazuo
drug discovery pegylated Christou, George
allograft cd4 endothelin
basic fibroblast base medicine effluent sars
sister chromatid phosphoinositide Burnett, John C
Tilney, N L
hepatocyte growth
Chen, Shih-Ann
antisense integrins
Mantzoros, Christos S kappab Aggarwal, Bharat B
phase transition coronary intervention single molecule
burgdorferi Myers, Martin G Low, John N
hydrogen bond raft amphiphilic
monoclonal trisphosphate gene delivery wastewater wall carbon
Nakamura, Toshikazu
polyarthritis catheter ablation care reform apoptotic cell her2 microarray
Bown, S G
tubercle bacillus adrenaline Häyry, P
color doppler transesophageal echocardiography adiponectin
photodynamic therapy propofol fatigue syndrome
Calkins, Hugh
leptin photocatalytic Glidewell, Christopher gastrointestinal stromal Kihara, Shinji
nmda
Wernsdorfer, Wolfgang
september symptomatology inquiry Blumberg, P M
orthotopic liver antiphospholipid syndrome thoracoscopic cd44 regulate kinase Taketo, Makoto M
electronic structure
fluoxetine prion
resonance image software synthesis characterization
Schiffrin, E L
lithotripsy
Davenport, A P
west nile Aguzzi, Adriano
egfr
therapeutic effect diaphragmatic birthday inoculate endobronchial metabotropic glutamate nurse practitioner bipolar disorder cd25
chain reaction signal transduction Herrlich, P
nucleotide polymorphism toll like
lanthanide Stang, Peter J Funahashi, Tohru
biologic
nitric oxide neural stem
Marks, F Wilske, B Cervera, R Prusiner, S B
legionella pneumophila zeolite
Wood, B P
Young, L W
personal experience
niche france Glick, P L
Tibboel, D
bufo possibility pathogenesis
Vale, W
adenovirus monoamine hodgkin lymphoma immunosorbent natural killer
intravenous immunoglobulin phorbol ester
Fikrig, E
helicobacter pylorus helicobacter workforce
Vieta, Eduard
Calabrese, Joseph R
total synthesis
Nicolaou, K C nile self assembly self assemble reader
Kim, Seung U observational study
tooth
Zewail, Ahmed H
tuberculous Blumberg, B S McCann, S M
phi annexin deficient mouse stereoselective ghrelin ultrafast quantum dot Pao, William
radiological
Miyamoto, E Graham, David Y Steinman, R M Seiki, M
ionize radiation
compute tomography
Perretti, Mauro
van de Beek, Diederik
vegetative streptokinase
tumor necrosis cd3 protein tyrosine icam Banchereau, J
transcriptome
meningitis
Corma, Avelino
theoretical study
Fürstner, Alois
intramolecular
Moretta, A
calmodulin dependent
endocrine orthodontic
Sinsheimer, R L Blaser, M J
nf kappab
Pestell, Richard G
Bechstein, W O
radiofrequency ablation pten 21st century multidetector h2o
italian
Bando, Yoshio
gene therapy
Armstrong, B K Alitalo, Kari O'Neill, Luke A J
mammary carcinoma hypoglycemicCuriel, David T
Doerfler, W
corticotropin aortic stenosis Cerhan, James R
macrophage colony
indication
Choi, DongilSpringer, T A Miyake, Kensuke
spontaneous pneumothorax
Thesleff, I
trh Herberman, R B Salas, M Noelle, R J
cd40
hormone release nanotube
Terhorst, C
immunodeficiency syndrome metabolic syndrome
Kavoussi, L R
australia
Cozen, Wendy Steinman, Ralph M
orthodontic treatment endothelial growth signal regulate Rotello, Vincent M
uranium immune complex relate quality mapk akt association study
Cademartiri, Filippo Gill, Inderbir S Kojima, Masayasu
pneumothorax
Gottlieb, E L Topol, E J
malpractice resveratrol
mitral stenosis meningeal cell therapy guerin
liver transplantation
Solcia, E Hosoda, Hiroshi
intestinal mucosa
laparoscopic desorption ionization club
Robbins, P D
McCracken, G H Gold, P W Oberg, K
appliance cardiac transplantation alpha fetoprotein list clinical medicine bromo
april remote asian
Pibarot, Philippe
echo polychlorinated biphenyl Arimura, A
endorphin forum cyclosporine subtraction
Brimacombe, J
reform human herpesvirus highly active catalyst Kobayashi, Shu Diamond, Michael S
like receptor
Grundy, Scott M Mirkin, Chad A
pneumoperitoneum
mycobacterium tuberculosis dentistry
Negri, E Chanock, Stephen J
assist laser
nanoparticle
Achenbach, Stephan
bcg
La Vecchia, C Stinson, E B
carcinoembryonic midazolam Oreopoulos, D G laryngeal mask single nucleotide
dentist genome sequence
Kowitz, A
mycoplasma adenylate cyclase ranitidine minimally invasive online toll
Auricchio, Angelo
antihistamine gynecological hyaluronidase Malendowicz, L K Storb, R
Misra, Anoop
prednisone amphotericin regime hong peer review
Robberecht, P
link immunosorbent Svehag, S E
refugee
Van Buren, C T
htlv Karin, M
shortage gabaa receptor mediate apoptosis privacy photonic climate change
Kubo, Michiaki Lapidus, Alla
pylorus
Bennett, W M
borrelia burgdorferi Kerr, Richard A
amp hybridoma coronary angioplasty edge resynchronization
systematic review
Wiesner, R H
campaign cor
Rastogi, Nalin
casualty Loevy, H T
massachusetts tetrahydrocannabinol Temin, H M Reis, Rui L
oncogene
Gessain, A
evidence base
Vacanti, J P
typhoid adrenal cortex aplastic anemia technetium 99m Bishop, J M Wherry, E John
dna vaccine cyclization
xii synergism soviet eeg immunosuppressive fish oil cd8 polyelectrolyte
Taylor-Robinson, D
argentina Jacobs, William R Collen, D styrene year result Iijima, Sumio
mtor pi3k Seo, Pil Ja
automate jun
Cooks, R Graham
Thomas, E D
isoniazid prednisolone imipramine Hardie, D G
bromocriptine zidovudine Ahmed, Rafi
chemokine receptor beta catenin Schalij, Martin J
Zhao, Dongyuan
experimental research
technic function test rheumatic disease
Kobayashi, G S
virus vaccine actinomycin immune deficiency tissue engineer photonic crystal spectrum disorder microrna
hiv
Job, C K
rubella kinase pathway dynamical
proteomic
Razin, S Budoff, Matthew J Tseng, L F
retrovirus
Schlom, J
shock wave proto oncogene oxy
pathological transaminase radioimmunoassay mesoporous
Higgins, C B
Bhatia, Kailash P
cell mediate
Muasher, S J Clotet, Bonaventura
map kinase ambulatory blood Ferrara, N
splice variant inducible nitric Butterfield, D Allan Ernst, Edzard
compute fluconazole
Jusko, W J
thrombolytic dentition cyclosporin tunable
Tobey, J A
legislation Brin, M F
echinococcosis hormone therapy charles leprosy ehrlich news Medoff, G
Hilleman, M R
dental caries
mycoplasma pneumoniae Sever, J L Saletu, B Chase, T N Yamada, K M
natriuretic factor
Hricak, H
yag laser
Rosenwaks, Z
proto kawasaki tat receptor gamma
Mancia, G
rapamycin aquaporin
Weiner, David B
electrospray ionization dendritic cell imatinib time pcr nanoscale Möhwald, Helmuth nanostructure
oxide synthase biomarker
Jerrold, Laurance
dystonia erythromycin reserpine haloperidol drug resistance poultry
Best, J M Pfurtscheller, G
levodopa
Clark, B F Schally, A V
125i Mosher, D F lymphocyte subset
yag patient safety millennium nile virus rituximab Fanti, Stefano
Shi, Jianlin
bevacizumab
pulmonary tuberculosis furosemide vitro fertilization
Jonas, Jost B
rheumatism cortisone defense
Ramu, G
lead poison triamcinolone MUHLER, J C elongation factor 99mtc somatostatin Li, C H
fibronectin captopril Gallo, R C
gene relate
simvastatin Pfaller, M A
tnf alpha mrsa Huang, P L
cyclin d1 diffusion weight severe acute Blennow, Kaj Pitluck, Sam
manage pet ct
Kantarjian, Hagop
butyl anabolic Oreland, Leung, Gabriel M
innate immune
Zetterberg, Henrik
1950
bronchial carcinoma orthodontics tanzania heart transplant Parmeggiani, A
cimetidine hdl acyclovir poverty octreotide
Moon, Randall T
amenorrhea kda
Youdim, M B Moncada, S Nishida, E
salmonella enterica
1960 1970
silicosis memoriam flow rate Lam, T H MacIntyre, I Sodroski, J
reverse transcription
1990
Richman, D D Rubello, Domenico
1980
Potter, B V Lee, Myeong Soo
L Houslay, M D Reubi, J C Ferrone, S Klein, T W Weissenstein, E Harris, A G Gisbert, J P
Khamashta, M A
gabaa maldi liquid crystal Holmes,
Aebersold, Ruedi
elevation myocardial benzyl personalize
2000
Soriano, Vincent
Kaminskaia, G O
internal medicine
VIDAL, J King, S B Nahorski, S R Maggi, C A Aggarwal, B B Wessely, S
Billiar, T R Agre, P Curiel, D T West, K W David R human embryonic iranian
Calin, George A
2010
puerperium 1951 1952 1953 1954 1955 1956 1957 1958 1959 1961 1962 1963 1964 1965 1966 1967 1968 1969 1971 1972 1973 1974 1975 1976 1977 1978 1979 1988 1991 1992
van Gunsteren, W F
Ng, Seik Weng Volinia, Stefano
graphene
typhoid fever b12
1981 1982 1989 1993 1994
surgical therapy physiopathology
1983 1984 1985 1986 1987 1995 1996 1997 1998 1999 2001 2002 2003 2004 2005 2006 2008 pluripotent
crown medicare sulfoxide
2007
Nussdorfer, G G
paris
dental pulp adoption
Conn, P M 2009 Tour, James M
cea methadone
Grimaldi, P L
professor Stitzer, M L vinyl
McCulloch, E A Kreek, M J
Hoppin, Jane A Magrath, I T
beta endorphin cftr
clinical aspect reactive protein cholesterol level burkitt Zamcheck, N Ziegler, J L Conn, P Jeffrey
allergic disease FLOCH, H laparotomy hong kong ambulatory peritoneal payment cytokine production Soderling, T R
release hormone calcitonin gene anti hiv 1beta
De Clercq, Erik
pathogenic tnf
Pannecouque, Christophe
blast layer chromatography
McEver, R P
process citation
agricultural triiodothyronine Chopra, I J glutamate receptor transition metal
Hogg, Robert S Ruoff, Rodney S
Bocci, A Tonks, N K
organic matter tio2
french civil relaxant anticholinergic
Ridker, Paul M
test result adriamycin computerize tomography zinc finger gemcitabine antiretroviral therapy genomics sirolimus solar cell
Lord, Catherine
layer
Soriano, V Lieberman, J A Prato, Maurizio
peculiarity family plan chloro
gastroduodenal korean york city rock Goldenberg, D M Amyes, S G prostacyclin jejuni Rivier, J release factor deficiency syndrome hiv infect immunodeficiency virus quantum meth
Sandler, Dale P Wainberg, M A Leon, Martin B Yamanaka, Shinya
dental practice
tubercle
nitric
Pfeiffer, L N
july Oppenheimer, J H Pepys, M B Thielemann, H
selectin
Grätzel, Michael
mitral laparoscopic surgery molecular dynamic zno
Strano, Michael S
open heart
Tremblay, Michel L
knockout mouse Cortes, Jorge mmp river
Miller, D Craig
dental
Hercberg, Serge
security white rat han Alfieri, Ottavio enterovirus body composition trimethoprim mediate cytotoxicity
Dang, C V
Peters, G J
Sorsa, T
hif silico
Katoh, Masaru
dft Fun, Hoong-Kun microbiota mir
disk hypothermia ketamine
Lee, Hyejung
address profound health plan
Gurney, B F
rifampicin observational Vane, J R legionella etoposide Howell, S B Gutkowska, J Cantin, M
docetaxel Tedder, T F p38
Ivanov, A Jang, Sung Ho
Jones, R N
president robert
Kim, Jong-Won Galea, Sandro Papakostas, George I
hyperbaric cyclic amp
Dumont, J E
acupuncture glycosylated
Greco, F A
quality improvement rt pcr Georgoulias, V
Andersson, Gerhard Barceló, Damià
collision initio ruthenium
Klionsky, Daniel J
h1n1 Mizushima, Noboru
leukotriene cell clone
p450 missensematrix assist internet induce apoptosis cell apoptosis gold nanoparticle McCammon, J Andrew
depressive dexamethasone
Sleytr, U B Stone, Gregg W
vietnam
human immunodeficiency dynamic simulation iran
Barry, A L Newman, Anne B
medicinal plant Ki, Chang-Seok
associate antigen biphenyl
focal
Harris, Tamara B
Herbert, V
hemorrhagic fever kong concanavalin adjuvant chemotherapy Edelstein, P H
nmda receptor clozapine
Knochel, Paul
diffusion tensor
tuberculous meningitis clinical result
anatomo belgium femoral neck hbsag sludge extracellular signal Katoh, Masuko
medico
meta analysis
Narayanan, P R Kimmel, K Waldman, H B Ford-Hutchinson, A W Gibson, C Michael
heart transplantation diagnostics
Shulman, S T
tetracycline serum cholesterol
Vlahov, David
dopa prostaglandin e2 enkephalin autophagy tetra
Motoyoshi, K Waksman, Ron
auxin
Dobryszycka, W Meyer, Helmut E
model base drug elute Maruoka, Keiji
partial denture
teeth buffalo monoclonal antibody ciprofloxacin ecosystem
Iannuzzi, L
massage transesophageal
tuberculosis
Wood, Robin
ultrasonic ussr prof
Misu, Y
presidential october myc amyloid precursor Beyreuther, K
bmp functionalized synuclein
extracorporeal sevoflurane
Fava, Maurizio Thompson, John F
cell leukemia cisplatinGolomb,
Eisenman, R N
polychlorinated paclitaxel beta1 alpha2 proteome single wall
Friml, Jirí Masters, C L Hollenberg, Morley D
atrial natriuretic
Ratcliffe, Peter J
hydrocortisone Golub, L M
Walle, T Gowda, B Thimme
Levy, S B apc
traditional chinese
Mori, Susumu
hyperbaric oxygen dental care root canal nurse care 17beta nuclear antigen beta2 sentinel genetic variant
mediate immunity erythropoietin
Thorsby, E
surface plasmon phenyl mass index
Kunkel, S L
hydatid cyst precede
Declerck, P J
chromatography tandem
invite adhesion molecule
Marsh, S G
chlorpromazine Saijo, N transduction pathway Ross, Merrick I Trost, Barry M Lee, Young Ho
ivf mr image
provoke König, K G
ct scan telomerase
hydroxydopamine fetoprotein
Chrambach, A extracorporeal circulation Bartlett, R H Ozsoylu, S
haptoglobin term memory dental health stellate cell enterica Cannon, Christopher P Ronco, Claudio
propranolol minimally
Coutinho, R A Baeuerle, P A
vitamin b12 dihydroxyvitamin pneumophila homosexual Samuelsson, B transluminal
HM
activator inhibitor
caries hla respiratory syndrome Montori, Victor M
acute kidney
infliximab biofilm chiral phen autism spectrum
preliminary note stock inositol polymerase chain Fisher, mycophenolate
Harvengt, C Lindner, Wolfgang
picture immunosorbent assay nf kappa
van Soolingen, Dick
tcr hcv Schachner, M
electrophoresis monoamine oxidase dependent kinase activate receptor
Taketa, K
claim
neoplasm lupus erythematosus
Deliens, Luc Tsuru, H Norman, A W Edelman, G M Calhoun, Vince D
wastewater treatment
accord czech denture chromatid exchange doppler echocardiography elute
DeLuca, H F
chloramphenicol actual carcinoembryonic antigen hospice hla dr acquire immunodeficiency Loskutoff, D J
antiphospholipid Shay, 1alpha
van Duijn, Cornelia M
thalidomide microsatellite
Guengerich, F Peter Peacock, Sharon J Narod, Steven A Levey, Andrew S
image guide
hiv infection
JW
Bowen, W H
affinity chromatography hla class sentinel lymph fmri target therapy supramolecular Knoll, Wolfgang
genome wide
transgenic mouse burkholderia mild cognitive
JW Meijer, E W
sedimentation fifty english Fehm, H L ferment eczema lipoprotein cholesterol
Feingold, M Goldwasser, E
campylobacter jejuni gamma interferon adenoviral
hypotension disease cause clinical picture catechol methylprednisolone proteomic analysis
Isenberg, D A Vandamme, Peter
Righetti, P G Mitchell, M D adjuvant therapy oximetry stat3 active antiretroviral reactor Schoepp, D D
coronary syndrome
Petersen, Ronald C
wave natriuretic density functional
Serruys, Patrick W
calmodulin
Werner, Liliana
antibiotic therapy Williams, H C Means, A R
nifedipine bulimia nd yag Zeuzem, Stefan
hiv type Shay, Jerry W
stat
radiologic acth cure thirty
Lubinski, Jan
especially igf
Montaner, Julio S G
Tunnessen, W W
scid
intraocular lens necrosis factor antiretroviral irinotecan
Madias, John E
metabotropic
p53 elute stent phen yl
Kawaoka, Yoshihiro
erythematosus
Hayaishi, O Shultz, L D Semenza, Gregg L Rivadeneira, Fernando
Yu, R K Brinster, R L
coleoptera
prostaglandin lepidoptera endothelial nitric biomass
Palmiter, R D
Webster, R G
amiodarone Mitchell, J E Staels, Bart
inducible factor pandemic
influenza
Ohman, E Magnus Fujimoto, James G
eosinophil blood protein animal experiment Apple, D J
gangliosides Guengerich, F P
Choi, Yung Hyun
o157 brca1 innate immunity ab initio coherence tomography Jove, Richard
Colombo, Antonio
wnt Levin, Adeera
anniversary Gleich, G J
hydrazide interferon gamma colony stimulate Metcalf, D
rflp Rosenfeld, R G pylorus infection Kroemer, G
atorvastatin clopidogrel Angiolillo, Dominick J Hoveyda, Amir H
Feng, Xiaoming
Venge, P
factor beta chronic fatigue LeRoith, D
high throughput excite state nano
apoptosis
Karch, H Guldi, Dirk M
st elevation chronic kidney
digestive multiple sclerosis autocrine Ungerstedt, U cytochrome p450
immunodeficiency
Oren, M
jnk
vegf olanzapine throughput
fullerene
acute respiratory ionic liquid
Gurbel, Paul A
Clevers, Hans Semenza, G L
gm csf Westerink, B H Lane, D P Komaroff, A L Ito, Osamu
mesenchymal stem
Miyazono, K Bellomo, Rinaldo
Miller, D H Filippi, M
cdna encode microdialysis Roberts, A B tyrosine phosphatase tgf beta cochlear implant enantioselective hypoxia inducible pcr assay Tohen, Mauricio
Alivisatos, A Paul
hmo
Tyler, R S
kawasaki disease
tgf telemedicinerisperidone apoptotic Krajewski, S intercellular adhesion
Tettamanti, G Letvin, N L Fujimoto, J G
Sporn, M B Klip, A
Merrell, Ronald C Antman, Elliott M
avian influenza carbon nanotube
catalyze
Creager, Mark A Ferrara, Napoleone
Traska, M R Kenkel, P J
granulocyte macrophage hiv positive glucose transporter
bcl Alitalo, K
wetland anthrax single center functional theory
Leppla, Stephen H
manage care pcr
Swayne, David E Rao, Mahendra S Nakatsuji, Norio
Pawson, T
mouse lack american college
Broxmeyer, H E
laparoscopic cholecystectomy snp microfluidic nanocrystal nanowire
Salvesen, Guy S
embryonic stem
percutaneous coronary
Doarn, Charles R
antiphospholipid antibody caspase Smith, Sidney C
web base phase microextraction
transform growth
Gelber, R D
Akerblom, H K
node biopsy ppar
cholecystectomy
Yang, Peidong
tyrosine kinase oxide production Vandenabeele, P tacrolimus Pawliszyn, Janusz
Mascitelli, Luca
expression profile
Capua, Ilaria
Devarajan, Prasad
kidney injury
study group press bax
Beller, Matthias Wang, Zhong Lin
inflammatory cytokine Korsmeyer, S J proteasome Strieter, R M chemokine statin
Kastelein, John J P
sirna coronavirus
Hughes, G R
cd34 web Dou, Q Ping interleukin 1beta coli o157 proliferator activate
Lai, M M
Wahli, Walter
groundwater
palladium
Figure 6: The graph G with year nodes, term nodes, author nodes and their
connections. We don’t show year-term and term-term connections, since there
are too many connections between these nodes.
6 Future improvements
Many techniques for automatic term recognition (ATR) are available in the
literature 4 . The approach for term extraction used in this study is very simple,
and a better approach could improve the results of this work.
For example, the terms human, immunodeficiency and virus are part of
a single multi-word term: ‘human immunodeficiency virus’. However, the
term immunodeficiency is extracted as a relevant term. ‘human’ and ‘virus’
are not recognized as relevant, since they are too broad. However, both ‘human
immunodeficiency’ and ‘immunodeficiency virus’ are extracted as relevant
terms. The full multi-word term is not extracted, since we limited the n-grams
7to a maximum of 2 words. We can find other examples of terms that are affected
by this problem, and it could be solved by using c-value for term extraction 5 ,
a method that combines linguistic and statistical information to extract multi-
word terms.
Another problem results from the choice of using lemmatization in the term
extraction phase. For example, the term ‘ionize radiation’ refers to ionizing
radiation, but the word ‘ionizing’ was lemmatized in the process, and was
maintained in this form for all the process to the visualization. This problem
could also be avoided by using better term extraction approaches.
Since this is an unsupervised method to extract terms and construct a visu-
alization, the method is susceptible to errors and noise in the data. For example,
the terms ‘july’ and ‘possibility’ were considered as relevant terms, even
if they can’t be associated to any discovery or breakthrough. A semi-supervised
methodology, where the help of a domain expert could be used to clean up the
data, would be preferred in this case.
Another way to validate the terms extracted by our approach is using the
UMLS vocabulary, which is also extracted from the MEDLINE R /PubMed R
dataset 6 . This vocabulary integrates over 730,000 biomedical concepts, and
some visual approaches were proposed for exploring semantic groups in this
data 7 . Such vocabulary can be used to verify the validity and relevance of
terms.
References
[1] R. P. Light, D. E. Polley, and K. Borner, “Open data and open code for big science
of science studies,” in Proceedings of International Society of Scientometrics and
Informetrics Conference, pp. 1342–1356, 2013.
[2] G. L. Rowe, S. A. Ambre, J. W. Burgoon, W. Ke, and K. Borner, “The scholarly
database and its utility for scientometrics research,” Scientometrics, vol. 79, May
2009.
[3] M. Bastian, S. Heymann, and M. Jacomy, “Gephi: An open source software for
exploring and manipulating networks.,” in ICWSM, The AAAI Press, 2009.
[4] K. Kageura and B. Umino, “Methods of automatic term recognition: A review,”
Terminology, vol. 3, no. 2, pp. 259–289, 1996.
[5] K. Frantzi, S. Ananiadou, and H. Mima, “Automatic recognition of multi-word
terms: the C-value/NC-value method,” International Journal on Digital Libraries,
vol. 3, no. 2, pp. 115–130, 2000.
[6] O. Bodenreider, “The Unified Medical Language System (UMLS): integrating
biomedical terminology,” Nucleic Acids Research, vol. 32, no. Database-Issue,
pp. 267–270, 2004.
[7] O. Bodenreider and A. T. McCray, “Exploring semantic groups through visual
approaches,” Journal of Biomedical Informatics, vol. 36, no. 6, pp. 414 – 432,
2003. Unified Medical Language System.
8You can also read
