Received 1 June 1981/Accepted 28 September 1981. The Hewlett-Packard HP 41C hand-held calculator can be used for the numerical identification of bacteria.
JOURNAL OF CLINICAL MICROBIOLOGY, Feb. 1982, p. 332-334 0095-1137/82/020332-03$02.00/0
Vol. 15, No. 2
Numerical Identification of Bacteria with a Hand-Held Calculator as an Alternative to Code Books JIRI SCHINDLER* AND ZDENEK SCHINDLER Department of Medical Microbiology and Immunology, Faculty of Medicine, Charles University, Prague, Czechoslovakia 12800
Received 1 June 1981/Accepted 28 September 1981
The Hewlett-Packard HP 41C hand-held calculator can be used for the numerical identification of bacteria. The dimensions of the identification matrix are limited to about 30 by 22; however, many groups of clinically important bacteria can be numerically identified by this method. Hand-held calculators can be used as an alternative to code books. At present, these calculators and additional tests can help solve identification problems in profiles not contained in code books.
Bacteria are identified by various tests. The media are either prepared in laboratories or, more frequently, supplied by various manufacturers as kits of microtests. The test results are interpreted mostly by code books supplied by manufacturers or by computers. Code books are supported by computer services which can be reached by phone 24 h a day in several instances. Computed results (4, 5, 7-9) can be obtained from desk-top computers (11) or personal computers. There are also automated devices that have appeared on the market (1, 6, 12, 13) for use in identifying bacteria. They work on different principles but use, however, computeraided identification for interpreting results. The aim of this paper is to show that handheld calculators can be used for evaluating test results. For economical reasons, diagnostic test kits are still widely used, and the results are interpreted by code books. However, computers and hand-held calculators offer more versatility than do test kits. Identification matrices containing frequencies of characters for different groups of organisms can be keyed in and stored on magnetic cards. A probabilistic algorithm (14) based on the Bayes theorem is used for numerical identification. We used the computing capabilities of a Hewlett-Packard HP 41C hand-held calculator with a total memory capacity of 255 storage registers alternatively connected with an HP 82143A thermal printer or an HP 82104A magnetic card reader. This small yet powerful computing system permits the management of large identification matrices with up to 660 elements. The printer provides a record of input data (strain number and test results) and corresponding identification (two taxa with computed identification scores) (Fig. 1). With the card reader, it is possible to load different identification matrices
stored on magnetic cards into the memory. The program itself occupies 52 registers. Five matrix elements are packed and stored into one register, with an accuracy of only two digits. This saving of memory space, however, helps to increase computation time. Great attention was given to ease of operation; to make print-outs understandable, we made full use of the alphanumeric capabilities of the system. To demonstrate the computing capability of the calculator in identifying bacteria, we chose the test set used in the Minitek code book Enterics. However, other test sets may be selected, or the number of tests used may be reduced. We assumed that the results in Enterics were computed with the data matrix contained there. For this reason, we used the same frequency matrix for evaluating our computations. This process would have taken about 200 h on the HP 41C, so all of the profiles in the code book were recalculated by the same algorithm written in HPL language and computed on the fast HP 9825 desk-top computer. The computation of one profile takes about a second. Thus, the 1,633 profiles contained in the code book could be computed in a relatively short time. In 16 of 1,633 profiles, the code book showed an identification score of P _ 0.9, and a score of P < 0.9 was calculated by the HP 41C, but both first-suggested choices were different. This means that in 0.98% of the profiles, the result of computation differed from the valid result of the code book. The difference in first-suggested choices, with a score of P < 0.9 in both the code book and the calculation, was observed in 160 profiles. Thus, computations by a hand-held calculator failed in less than 1% of the profiles in the code book. This difference was caused by the fact that values rounded to two decimal places were used in the calculations. In certain 332
NOTES
VOL. 15, 1982
2568*A;
1j + 92F or*t + 7, Y
+
Ifl81- *J
1.a dlJ
8 a xyr
U
E 4-53 C_ Co L SP ENT
t sF) } or
a3r hi-.-.
t _
i,I X r
4.4 I
bi *+Je ; * t +
-2 R
so:
-l . 1355 0
9
Ci - a H ; '_ U
~pical
FIG. 1. Print-out of a ty identification result with an HP 41C. The isollate number is 2568 *** Abbreviations: ni, nitrates; ph, phenylalanine deaminase; ci, Simmons citrate; on, o-nitrophenyl-o-D-galactopyranoside; or, ornithiine decarboxylase; ly, lysine decarboxylase; ar, arabinose; ml, malonate; rh, rhamnose; id, indol; h2, H2S ; ur, urea; in, inositol; vp, acetoin; ag, arginine dihy(drolase; rf, rafinose, du, dulcitol; xy, xylose; so, sorb itol; la, lactose; ad, adonitol; sa, sucrose; +, positive; ES COL, Escherichia coli; ;SASlenteritidis.
EgNTlve; notnelladone
critical instances, this iinfluences the resulting scores and thus, influeinces the identification results. Total agreement of thie results computed by the HP 41C has been obitained in a test battery and code book describe d elsewhere (10). The code book of 10 tests contained about 1,000 profiles. The identification matrix used for the construction of the code book was derived from Ewing tables (2). Agreement was also obtained with an API 10S code book. The identification matrix was derived from an API information sheet and contained 10 tests and 33 species. The identification of viridans streptococci, which can still be a problem, may be performed by computation with data published by Facklam (3). A total of 25 tests were selected to identify 10 proposed species of streptococci. The identification matrix contained 25 tests and 10 taxa. The identification of streptococci served as an example of the direct computing of test results, thus creating, in fact, a dynamic code book. The computing time in this case was about 2.5 min. HP 41C can be used for numerical identification, but its use is limited by memory capacity. However, the limiting dimensions of the matrix (about 660 elements) allowed the identification of many groups
333
of bacteria encountered in clinical microbiology. A second limitation of the HP 41C is only of a relative nature: the computation using the maximal matrix takes 5 to 7 min. However, in the routine work of most laboratories, only a small proportion of strains needs to be identified numerically because of rare test patterns. Usually, a set of 10 to 12 tests is satisfactory for the identification of frequently occurring patterns (10). Unusual strains need to be examined by an extended set of tests and interpreted by a voluminous code book. For the majority of strains, hand-held calculators offer a plausible alternative to code books. The capacity and ratio of price to performance of a new class of hand-held calculators represented by the HP 41C enable laboratory workers to compute their own results. Personal computing has several advantages over code books. The versatility of the laboratory-prepared tests of the Minitek system allows workers to design sets of tests and, thus, to define identification systems supported by numerical identification. Another advantage is accuracy. The results of additional tests recommended by code books sometimes do not discriminate between the suggested species of particular profiles. In these instances, computed results may approach the correct results. Personal computing also allows for immediate consultation. If the code book fails, the isolated strain can be identified by computing the results of all of the tests available. The numerical identification of bacteria, based on the Minitek test kit results, complements the present Minitek identification system. These calculations can be performed in every laboratory with a hand-held calculator that costs about the same as three code books. LITERATURE CITED 1. Brenner, D. J., and A. Balows. 1975. Evaluation of the Enteric Analyzer, an instrument to aid in the identification of Enterobacteriaceae. J. Clin. Microbiol. 2:235-242. 2. Ewing, W. J. 1973. Differentiation of Enterobacteriaceae by biochemical reactions. U.S. Department of Health, Education and Welfare, Atlanta. 3. Facklam, R. R. 1977. Physiological differentiation of viridans streptococci. J. Clin. Microbiol. 5:184-201. 4. Friedman, R., D. Bruce, J. MacLowry, and V. Brenner. 1973. Computer-assisted identification of bacteria. Am. J. Clin. Pathol. 60:395-403. 5. Gyllenberg, H. G. 1965. A model for computer identification of micro-organisms. J. Gen. Microbiol. 39:401-405. 6. Isenberg, H. D., T. L. Gavan, P. B. Smith, A. Sonnenwirth, W. Taylor, W. J. Martin, D. Rhoden, and A. Balows. 1980. Collaborative investigation of the AutoMicrobic System Enterobacteriaceae biochemical card. J. Clin. Microbiol. 11:694-702. 7. Lapage, S. P., S. Bascomb, W. R. Willcox, and M. A. Curtiss. 1973. Identification of bacteria by computer: general aspects and perspectives. J. Gen. Microbiol. 77:273-290. 8. MacLowry, J. D., and E. A. Robertson. 1978. Applications of computer diagnostic models in the clinical micro-
334
NOTES
biology laboratory, p. 293-318. In A. Sharpe and D. S. Clark (ed.), Mechanizing microbiology. Charles C Thomas, Publisher, Springfield, Ill. 9. Rypka, E. W., W. E. Clapper, I. G. Bowen, and R. Babb. 1967. A model for the identification of bacteria. J. Gen. Microbiol. 46:407-424. 10. Schindler, J., J. Duben, 0. Hausner, V. Zikmundovl, J. Lapakova, and V. Pau&kovd. 1980. Standard test set and a computer-generated diagnostic register for gram-negative fermenting rods. J. Appl. Bacteriol. 49:331-337. 11. Schindler, J., J. Duben, and 0. Lysenko. 1979. Computeraided numerical identification of gram-negative fermenta-
J. CLIN. MICROBIOL. tive rods on a desk-top computer. J. Appl. Bacteriol. 47:45-51. 12. Shayegani, M., M. E. Hubbard, T. Hiscott, D. M. McGlynn, and R. C. Yewdall. 1975. Evaluation of the Enteric Analyzer for identification of Enterobacteriaceae. J. Clin. Microbiol. 2:186-192. 13. Sonnenwirth, A. C. 1977. Preprototype of an automated microbial detection and identification system: a developmental investigation. J. Clin. Microbiol. 6:400-405. 14. Willcox, W. R., S. P. Lapage, S. Bascomb, and M. A. Curtiss. 1973. Identification of bacteria by computer. Theory and programming. J. Gen. Microbiol. 77:317-330.
