Clive—

June 12, 1956

Pi bruce, et alf
vustg got your tape; 4ere's an impromptu reply.

still straggling with the ms, I've decided to follow you on terminology
and forget about cetenate. 80 I'm back to line (emph. uni-linear vs. milti-
linsar) gor cuain, and so forth, fo new data at all, and I ope not to do
any more experiments on thie for a while, Only interetsting thing cleaned
cut af ay notes that i haven't already told you is a pedigree in which
a sib to a ewerm gave 22 lines. The boundary between E and non-E will have
to be rather arbitrary. Still hope to have this in a 1354 Genetics, Haven't
actually done any experiments myself om lines since last summer, Did I
tell you then that wotile lines isolated at an intermediate generation
fro Flat Hy" —x SU666 were xitbchcticttent not all inkihited by anti- a a
thuugi git the initiesis were at least partially inhinicved, Goes this agree
with you Ynow (I hope I have thet straight; don't have my notes here this
PM). I burned aw my fingers with the i—x &W-666 experiments, as the anti- i
serum algo inkihited b-—x SW666, much to my surprise. (This effect mly am
trails and isolated initials, not in agglutination or swarms; I hope you have
such a control and have same idea what the reaction 4s, The serum of course
is Colindale's).

With n binary tests of course 2" classes should be distinguishable, but
Bruce is right that you ought to leave out 000000 and 111211 for confirmation.
We used to rundown nutritional requirements of auxotrophs that way, but the
trouble was that each unit reaction has te be perfect for the canpound to
be reliavle, and I would ve very leary of relying on Salmonella agglutinations
so trustingly. I suppose the method sheuld be ok for preliminary screening,
but I feel you should have a redundancy of information in practical typing.
When you serotype with individual, or with pooled reagents in d&screte groups,
you know that you must get a code like 00001000 and that 01000100 means something
is wrong, One can, of course compromise between maximm information and maxi-
mm semurity, but i suspect that considerations like this have discouraged
earlier applications. Do you think you can get away with a conpletely effi-
elent code for Salaciella typing? By the way, a binary numbering scheme should
make it easier to <<, up and to translate the efficient codes; as I suppose
you've worked out in ome form or another, Pool A should define the first digit
(e.¢., for 16 rearents should contain # 1-8), pool B the second (#1-4, 9-12
pool G the 3d (1-2, 5G, 9-10, etc) and pool D the 4th (odd #!a), Then a
code like 0101 is readily translated as #5, excopt 1 huve it backwards and
O=+, 1 =~ reaction, I found this type of mmbering very handy in sumerizing
genotypes; after translating to decimal, it's mich easier te scan for how many
type 7's than for how many BXXXM{XaEAY ~+44+'s in a table full of + and ts,

I'll be interested to hear hoy this works owt in practice,