Transcript A Tale of Three Inferences

A Tale of Three Inferences
Models of transcription and their
consequences
Philip Benfey, Ken Birnbaum, Dennis
Shasha
What is the Logic of
Transcription?
• Known: transcription factors bind to small
subsequences of DNA, perhaps in a
statistical mechanical (hence concentration
dependent) way.
• Controversial: interaction among different
transcription factor-binding events.
Models of Interaction
• Additive (Boolean OR): For promoter P on
gene G, if T1 binds to c1 and T2 binds to c2
in an inductive way, then the expression of
G will remain the same if the promoter were
to have twice the number of c1 and c2 goes
to 0.
• Boolean AND: Under same conditions,
there will be no expression
Classical Approach
• Detailed nature of interaction is unknown.
• Find genes whose expressions correlate well
with one another.
• Infer that common motifs among the
promoters of those genes must be the
binding cis-elements.
Note to ken
• Ken, please include a power point slide in
which you show how this correlation
method is meant to work.
Bussemaker, Siggia, et. al
• Model: expression of a promoter is:
 S coef * number, where coef[j] is the
strength of the jth transcription factor-ciselement binding and number[j] is the
number of times the cis-element appears.
• Additive, Boolean disjunctive model.
• No explicit model of transcription factors.
Bussemaker Discovery
• Consider one experiment consisting of
many gene-promoter pairs some of which
are expressed.
• Best cis-element (largest coefficient) is one
that appears in promoters of many
expressed genes and few unexpressed genes
(inversely weighted by number of
appearances per promoter)
Bussemaker Example
•
•
•
•
Promoter 1: x, x, x, y. Expression: 1
Promoter 2: x, y. Expression 1.
Promoter 3: x, y, y, y. Expression 3.
Conclude: best hypothesis is that each y
contributes to expression with a coefficient
of 1. No good model for x.
NYU (Birn,Benf,Sha) Model
• Transcription factor/cis-elements form a
boolean (AND/OR/NOT) model with
amplifiers (e.g. TFA & TFB ==> TFC)
• Finding these circuits is a multi-stage affair:
1) find cis-element/transcription factor pairs
2) infer boolean circuit by seeing
combinations that work.
NYU Find TF/Cis-element pairs
• Given the knowledge of genes that encode
transcription factors and given a sequence
of experiments (e.g. different times during
sporulation), find cis-elements that correlate
best with transcription factors over time.
NYU Model Details
• Cis-element c expression at time t = sum of
expression of genes containing c in their
promoter sequence.
• Transcription X expression at time t =
extent of RNA production at time t.
• X binds to c potentially if two time
sequences correlate well.
Note to Ken
• Ken, please put in our experiment with
promoters A, B, C such that first promoter
has AB, second has AC, and third has BC.
First time has TA, TB; second time has TA,
TC; and third time has TB, TC.
• Gene correlation doesn’t work.
• Bussemaker only partly works.