Information Content of a Human Being
---Warren D. Smith May 2001---------

The human genome is 3*10^9 base pairs (bp) i.e. 6 gigabits.
However, really, most of this is "junk."

There were fewer human genes than expected, because initially
the number of genes had been estimated based on a segment of DNA that was
anomalously gene-rich.  Hence there is more junk than expected.
Craig Ventner now (April 2001) claims he estimates there are 26400 genes
with at least 2 supporting lines of evidence, plus about 12000 additional
gene candidates, most of which (he suspects) will turn out not to be real.
The Human Genome is discussed in Science 291,5507 (16 Feb 2001) and
Nature 409,6822 (15 Feb 2001), both special issues.

creature
--------                       #genes  #bp
Human                          32000?   3*10^9
Fly (Drosophila Melanogaster)  13338
worm (C.Elegans)               18266
Yeast                           6144
Mustard Weed                   25706
H.Influenzae                    1743   1830137

Two random humans differ in 1 base pair out of 1250.
That implies that the amount of information needed to describe how you 
genetically DIFFER from me, is only about  50 megabits   at most.
Of course this is probably an overestimate since most of those
differences actually have no discernible effect.

It is suspected that if all 6 billion human genomes were
sequenced, then almost all possible point mutations would be seen.
It has been estimated that 1-100 point mutations occur per human generation
going back 5000 generations. (I personally have estimated: 60.)

The average span of a "typical" human gene is 27894 bp
and it has 7 exons. The largest known human gene is for titin with
234 exons and including about 89 kbp of genuine coding DNA.
Chromosome 19 has 23 genes per Mbp (max), Chr. 3 has only 5 (min).
The % of human bp spanned by genes is 25.5-37.8%. The % of bp that are in exons
(i.e. code for protein) is 1.1-1.4%. The % that is introns (i.e. transcribed
to RNA but discarded) is 24.4-36.4%. The % that is intergenic is 63.6-74.5%.
The longest known intergenic length is 3038416 bp.
Four classes of "parasitic elements"make up 45% of your DNA; most of
this arose from reverse transcription of RNA. 
3% of your DNA is short repeats.
5% is duplicated large pieces.

The true number of genes is not known but it won't change
by a heck of a lot. It is for sure between 26000 and 39000.
There are also promoters, repressors, and introns, and I don't
think they should count as "junk," at least not 100% junk. So
maybe human DNA is 98-99% junk.

Now, even of the 1-2% non-junk, it could be argued that in fact,
a good deal of it is not really transmitting information, since,
if I change a random non-junk bp it probably will usually have no effect
on your life. I asked Ned Wingreen once about that & he estimated
that a non-junk DNA bp (actually a random bacterial bp)
would, if changed, cause 2-3% probability of harm.  

In that case the "true info content" of human DNA is more like 2-3% of 1%,
which is .0003 fraction, say, which for humans is 9*10^5 bp which
is 1.8*10^6 bits. That is comparable to the amount of info in
a typical paperback book, I guess.

On the other hand, these estimates of harm caused by changing just ONE bp
in isolation may tell us little about true info content - I suspect 
randomizing half of your gene bp's would kill you, even if you got to
select the least harmful half before I flipped my coins! If we guess that
maybe 1/6 of the bits in gene DNA actually really do matter
(6 bits = 1 triplet specify an amino; 1 bit specifies
hydrophobic vs hydrophilic) which would be 16.6% not 2-3%.
In that case the "true info content" of human DNA is more like .0016 to
.0032 fractionally, i.e., 5*10^6 to 10^7 bp, i.e.:
  10-20 megabits.

Another estimate of something similar:
[AD Keefe & JW Szostak: Functional proteins from a random sequence library,
Nature 410 (5 Apr 2001) 715]  estimates (based on an experiment) that
among 80-amino proteins with random sequences, about 1 in 10^11 of them will 
"work," i.e. have a chemical functionality you specify.
(Their experiment with 6*10^12 such proteins found 4 that worked,
i.e. had ATP-binding functionality.)

That suggests that the true "entropy" of a gene is
about log2(10^11) = 36.5 bits per gene.
If we want the thing not only to work, but also work well,
perhaps that brings the entropy up to 70 bits per gene
(twice as much; this is a guess). 
Anyway, this is an interesting number.

If humans have 30000 genes that means our entropy, in some sense,
is only 2.1 Mbits. This is close to the previous estimate based
onm Wingreen's guess.

---------------------------------

Let's contrast this with the amount of mental information you learn
and store in your brain over your lifetime.
Psychological experiments have estimated [T.K.Landauer: 
How much do people remember, Cognitive Sci. 10,4 (1986) 477-493;
12,2 (1988) 293-297] that you remember about 2 bits per second
and end up holding somewhere between .5 and 3.4 gigabits.  This is smaller
than the junk-info content of your genome, but a lot bigger than its true-info
content.