MetaMap Indexing Algorithm

Indexing Initiative
The MetaMap Indexing (MMI) methodology consists of: The rest of these notes consist of a description of the MMI ranking function followed by some examples.

Definition of the ranking function:

Generally speaking, the MMI ranking function was designed to indicate the characterizing power or aboutness of a given concept for a piece of text, e.g., a MEDLINE® citation. It is the product of a frequency factor and a relevance factor. The relevance factor is, in turn, a weighted average of four components (listed in order of importance): a MeSH tree depth factor, a word length factor, a character count factor, and a MetaMap score factor. For concepts found in the title of the citation, there is a simplified form of the function which maximizes the frequency factor.

The components of the function:

Before definining ranking function components, themselves, it id worthwhile to point out that each component is normalized before being combined with other function components. Normalization is accomplished by applying a sigmoidal function to a value. Each sigmoidal function maps from the unit interval [0,1] to itself and is determined by an index. An index of zero determines the identity function (hence, no normalization). A positive index determines a sigmoidal function which lies above the identity and consequently accentuates the differences between small values. Larger indexes produce more accentuation. Conversely a negative index determines a sigmoidal function lying below the identity and consequently accentuates the differences between large values. If we denote the normalization function with index n by nu n :[0,1] -> [0,1], then the formal definition of them is:

nu 0 (x) = x;
for n > 0,
Formula 1 image
for n < 0 (letting m=-n),
Formula 2 image
Now the basic ranking function components are defined as follows:
  • the frequency, simply denoted f, is the number of times the concept occurs (i.e., is discovered by MetaMap) in the title or abstract fields of the citation divided by 13;

  • the MeSH tree depth, m, is the maximum depth of all MeSH tree codes associated with the concept divided by 9. The computation is actually done by counting the number of periods in the tree code;

  • the word length, denoted w, is the number of words in the concept divided by 26;

  • the character count, denoted c, is the number of characters in the concept divided by 102; and

  • the MetaMap score, denoted mm, is just the score divided by 1,000.
Putting it all together:

Using weighting factors denoted by w m , w w , w c, w mm for MeSH tree depth, word count, character count, and MetaMap score, respectively, the ranking score can be written as:
Formula 3 image

The actual function with weights determined by experimentation is:
Formula 4 image

When the concept occurs in the title of the citation, the frequency factor nu f (f) is replaced by 1. This has the effect of giving title concepts overwhelmingly good rankings.

Ranking examples:

This section contains five examples with 3-point average precisions ranging from superior to the worst possible. Each example consists of the MEDLINE citation and the ranked MMI output. An exclamation mark indicates that the MeSH heading found by MMI occurs in the MH field of the citation; an asterisk indicates a main heading.

The best:

An example with near perfect 3-point average precision of 0.9496.

UI - 93121691
AU - Kario K ; Matsuo T ; Nakao K
TI - Cigarette smoking increases the mean platelet volume in elderly
     patients with risk factors for atherosclerosis.
AB - To study the effects of cigarette smoking and atherosclerosis on 
     platelet size, we measured the mean platelet volume (MPV) and other
     platelet parameters in 142 elderly smokers and nonsmokers with or
     without atherosclerotic risk factors. The MPV and the platelet count
     were highest and their inverse correlation was strongest in the
     atherosclerotic smokers (r = 0.54, P < 0.05) when compared
     with the nonsmoking and non-atherosclerotic groups. A 10% decrease of
     MPV was found in 8 smoking subjects in the atherosclerotic group, who
     successfully discontinued smoking (P < 0.05). These results suggest
     that smoking may increase platelet consumption in atherosclerotic vessels
     and that subsequently megakaryocytes are activated to produce larger
     platelets, which are more active. Thus, an increase in MPV due to smoking
     may also contribute to the acceleration of atherosclerosis and should be
     considered as a risk factor for atherosclerotic disease.
MH - Aged ; Aged, 80 and over ; Atherosclerosis/*BLOOD ; 
     Blood Platelets/ *ULTRASTRUCTURE ; Cell Size ; Comparative Study ;
     Female ; Hematopoiesis ; Human ; Male ; Megakaryocytes/CYTOLOGY ;
     Platelet Count ; Risk Factors ; Smoking/*BLOOD ; 
     Support, Non-U.S. Gov't
SO - Clin Lab Haematol 1992;14(4):281-7
PY - 2
MMI concepts for 93121691 (3-pt AP=0.949603):
! 59.5 Risk Factors
! 48.2 *Atherosclerosis
! 34.1 *Blood Platelets
! 33.8 Aged
! 25.8 *Smoking
17.4 Patients
! 9.3 Platelet Count
7.7 Acceleration
! 6.6 Megakaryocytes
5.0 Disease


An example with good 3-point average precision of 0.8808.

UI - 93230262
AU - Bellemare F
TI - Evaluation of human diaphragm function.
AB - When single supramaximal shocks are delivered during relaxation to both
     phrenic nerves simultaneously, the resulting transdiaphragmatic pressure
     twitch (PdiT) or mouth pressure twitch (PmT) are found to decrease linearly
     with increasing lung volume thereby reflecting changes in diaphragm
     contractility. Whereas fatigue decreases PdiT at any given lung volume,
     chronic lung hyperinflation tends to increase PdiT at any given lung
     volume. When the phrenic nerve shocks are delivered during ongoing
     voluntary contractions, PdiT decreases with increasing level of diaphragm
     activation. Its amplitude thus detects the reverse left for full activation
     of the diaphragm by the voluntary motor drive. The ability to maximally
     activate the diaphragm decreases with fatigue but is retained in patients
     with chronic lung hyperinflation.
MH - Diaphragm/*PHYSIOLOGY; Human ; 
     Lung Diseases, Obstructive/ PHYSIOPATHOLOGY ; 
     Muscle Relaxation/PHYSIOLOGY; Phrenic Nerve/ PHYSIOLOGY
SO - Monaldi Arch Chest Dis 1993;48(1):92-3
PY - 3
MMI concepts for 93230262 (3-pt AP=0.880795):
! 42.3 *Diaphragm
! 18.9 Phrenic Nerve
9.5 Fatigue
9.5 Shock
9.5 Pressure
6.5 Mouth
6.4 Lung
6.1 Relaxation
5.0 Social Change
4.9 Drive
4.5 Patients


An example with moderate 3-point average precision of 0.4867.

UI - 93051935
AU - Vermerie N ; Kusielewicz D ; Tod M ; Nicolas P ; Perret G ; Fauvelle F ;
     Petitjean O
TI - Pharmacokinetics of glafenine and glafenic acid in patients with cirrhosis,
     compared to healthy volunteers.
AB - Pharmacokinetic parameters were evaluated in 12 patients with alcoholic
     cirrhosis and 12 healthy volunteers after a single 400 mg oral dose of
     glafenine. Glafenine (G) and its major active metabolite glafenic acid (GA)
     were measured at regular intervals using a specific high performance liquid
     chromatographic method. Glafenine absorption was significantly delayed in
     cirrhotic patients (CP) (Tmax = 2.8 +/- 1.3 hvs 1.5 +/- 0.4 h, p less than
     0.01) and was dramatically reduced in 3 patients. The large hepatic 'first
     pass' effect observed in healthy volunteers was markedly reduced in CP
     (ratio Cmax GA/Cmax G = 3.6 +/- 2.9 vs 18.9 +/- 9.8, p less than 0.001;
     ratio areas under the curves AUC GA/AUC G = 2.3 +/- 2.3 vs 18.2 +/- 11.2,
     p less than 0.001). The elimination half-life of G was prolonged in the CP
     (13.0 +/- 13.1 h vs 1.5 +/- 0.5 h, p less than 0.01). In CP, GA elimination
     half-life was increased (12.0 +/- 13.4 h vs 4.3 +/- 1.3 h, NS) but the
     difference did not reach statistical significance because of large
     variability. The significant rise of G plasma concentrations (Cmax = 2.2
     +/- 2.1 mg/L vs 0.7 +/- 0.2 mg/L, p less than 0.05) and its longer
     half-life would lead to an accumulation if the usual dosage regimen was
     prescribed for CP and could result in nephrotoxicity. On the other hand,
     lower dosage would be ineffective because only GA is active and
     nephrotoxic. Hence, G should be given with great caution to CP.
MH - Administration, Oral; Adult ; Comparative Study ; Female ;
     Liver Cirrhosis, Alcoholic/DRUG THERAPY/*METABOLISM ; Male ; Middle Age
SO - Fundam Clin Pharmacol 1992;6(4-5):197-203
PY - 2
MMI concepts for 93051935 (3-pt AP=0.486722):
! 59.0 *Glafenine
31.5 Pharmacokinetics
17.6 Voluntary Workers
17.5 Patients
13.7 Half-Life
13.0 Gases
! 10.9 *Liver Cirrhosis,
6.6 Absorption
6.6 Hand
6.5 Plasma
6.5 Mouth
4.9 Lead
4.9 Methods


An example with poor 3-point average precision of 0.2688.

UI - 93287215
AU - Thali M ; Moore JP; Furman C ; Charles M ; Ho DD ; Robinson J ; Sodroski J
TI - Characterization of conserved human immunodeficiency virus type 1 gp120
     neutralization epitopes exposed upon gp120-CD4 binding.
AB - Interaction with the CD4 receptor enhances the exposure on the human
     immunodeficiency type 1 gp120 exterior envelope glycoprotein of conserved,
     conformation-dependent epitopes recognized by the 17b and 48d neutralizing
     monoclonal antibodies. The 17b and 48d antibodies compete with anti-CD4
     binding antibodies such as 15e or 21h, which recognize discontinuous gp120
     sequences near the CD4 binding region. To characterize the 17b and 48d
     epitopes, a panel of human immunodeficiency virus type 1 gp120 mutants was
     tested for recognition by these antibodies in the absence or presence of
     soluble CD4. Single amino acid changes in five discontinuous, conserved,
     and generally hydrophobic regions of the gp120 glycoprotein resulted in
     decreased recognition and neutralization by the 17b and 48d antibodies.
     Some of these regions overlap those previously shown to be important for
     binding of the 15e and 21h antibodies or for CD4 binding. These results
     suggest that discontinuous, conserved epitopes proximal to the binding
     sites for both CD4 and anti-CD4 binding antibodies become better exposed
     upon CD4 binding and can serve as targets for neutralizing antibodies.
MH - Amino Acid Sequence ; Animal ; Antibodies, Monoclonal/IMMUNOLOGY ;
     Epitopes ; Antigens, CD4/*METABOLISM ; Cell Line ; Cercopithecus aethiops ;
     Human ; HIV Antigens/*IMMUNOLOGY ; HIV Envelope Protein gp120/*IMMUNOLOGY ;
     Macromolecular Systems ; Molecular Sequence Data ; Neutralization Tests ;
     Protein Conformation ; Receptors, Virus/ *METABOLISM ; 
     Recombinant Proteins/IMMUNOLOGY ; Structure-Activity Relationship ;
     Support, Non-U.S. Gov't ; Support, U.S. Gov't, P.H.S.
SO - J Virol 1993 Jul;67(7):3978-88
PY - 2
MMI concepts for 93287215 (3-pt AP=0.268771):
67.3 HIV-1
! 34.0 Epitopes
19.5 Antibodies
12.8 Antibodies, Anti-Idiotypic
12.7 Glycoproteins
! 11.5 *Antigens, CD4
8.1 Rabies
! 7.7 Antibodies, Monoclonal
6.6 Molecular Conformation
6.1 Binding Sites
5.0 Personality Tests
5.0 Social Change
3.4 Immunologic Deficiency Syndromes
3.3 Amino Acids

The Worst:

An example with universally bad precision of 0.0. The 3-point average is, of course, also 0.0.

UI - 93020241
AU - Li WS ; McChesney JD
TI - Preparation of potential anti-inflammatory agents from dehydroabietic acid.
AB - Methyl 16-nor-16-carboxydehydroabietate (22),
     16-nor-16-carboxydehydroabietinol acetate (23), methyl 7-keto-16-nor-16-
     carboxydehydroabietate (29), 16-nor-16-carboxydehydroabietic acid (30),
     16-nor-16-carboxydehydroabietinol (31), 7-keto-16-nor-16-
     carboxydehydroabietic acid (32), methyl 7-hydroxy-16-nor-16-
     carboxydehydroabietate (33), and 7-hydroxy-16-nor-16-carboxydehydroabietic
     acid (34) were prepared from dehydroabietic acid. Only 22 and 32 had weak
     anti-inflammatory activity.
MH - Anti-Inflammatory Agents, Non-Steroidal/CHEMISTRY/*ISOLATION &
SO - J Pharm Sci 1992 Jul;81(7):646-51
PY - 2
MMI concepts for 93020241 (3-pt AP=0.000000):
17.7 Anti-Inflammatory Agents
9.1 Acids
8.2 Acetates

Last Modified: May 30, 2019 ii-public2
     Contact Us    |   Contact Us (SemRep)    |   Copyright    |   Privacy    |   Accessibility    |   Freedom of Information Act    |    Get Acrobat Reader button
Links to Our Sites:
Indexing Initiative (II)
Investigating computer-assisted and fully automatic methodologies for indexing biomedical text. Includes the NLM Medical Text Indexer (MTI).
Semantic Knowledge Representation (SKR)
Develop programs to provide usable semantic representation of biomedical text. Includes the SemRep program.
Program to map biomedical text to the UMLS Metathesaurus. Information and downloadable material for the MetaMap program.
Word Sense Disambiguation (WSD)
Test collection of manually curated MetaMap ambiguity resolution in support of word sense disambiguation research.
MEDLINE Baseline Repository (MBR)
Static MEDLINE® Baselines for use in research involving biomedical citations. Allows for query searches and test collection creation.
Structured Abstracts (SA)
Information about NLM's research on Structured Abstracts in the MEDLINE® Baselines.
Lister Hill Center Homepage Link - Image of Lister Hill Center Lister Hill National Center for Biomedical Communications   NLM Homepage Link - NLM Logo U.S. National Library of Medicine   NIH Homepage Link - NIH Logo National Institutes of Health
DHHS Homepage Link - DHHS Logo Department of Health and Human Services